Claude Code is steganographically marking requests

https://thereallo.dev/blog/claude-code-prompt-steganography

by kirushik • 22 days ago

2445 677 22 days ago

LPisGood• 22 days ago

This is very interesting. Combating resellers and distillation seems like a very difficult problem indeed. Interesting to me is that these techniques mentioned in the article are just like anti-observation techniques used by some of the more sophisticated malware out there, however defeating them is pretty trivial.

mysterydip• 22 days ago

seems ironically like a similar problem of content owners trying to filter bot scrapers from legit users

_alternator_3• 22 days ago

Yes, defeating this is relatively easy, particularly for sophisticated actors. But it's hard to always defeat all of the tricks. Sort of like how it's expensive and hard and uncertain to defeat all of the tricks when forging money.

Here's an example. Say you have your team use patched binaries. Then CC updates and requires a new patched binary with new tricks. You now have to have a team ready to analyze the binary and begin to address the tricks; meanwhile, unpatched code is now a fingerprint. If some researcher decides to update Claude on their own to access new features, they get fingerprinted.

Defeating a single fingerprinting technique once is easy. Defeating all of the techniques all the time is hard.

charcircuit3• 22 days ago

Is it hard? Just ask AI if the update added any new fingerprinting vectors?

SubiculumCode5• 22 days ago

Not to mention, it isn't that hard for vendor's to require updated code to run the product. Vendors do this all the time.

pishpash4• 22 days ago

Corporate surveillance malware on employee machines is also defeatable but most don't bother.

Laurel12347• 22 days ago

these techniques mentioned in the article are just like anti-observation techniques used by some of the more sophisticated malware out there, however defeating them is pretty trivial.

Really makes you think huh

MattDamonSpace• 22 days ago

“So the feature mostly punishes the exact people who are easier to fingerprint: normal developers doing weird but legitimate things”

What’s the punishment here exactly?

pedropaulovc• 22 days ago

Higher odds of being banned for legitimate usage.

solenoid093718• 22 days ago

If you are accessing Claude through the listed domains it is not "legitimate use."

femboyvtuber• 22 days ago

Returning invalid poisoned different results that were not what you paid for

bakugo3• 22 days ago

Output poisoning and/or eventual account bans, if I had to guess.

realusername• 22 days ago

They probably run a heavily dumbed down version of the model, same as what they got caught doing with Fable.

And that's also why, as a legitimate customer, want none of it, you never know if you accidentally entered a zone they don't like.

mgraczyk• 22 days ago

"got caught"

to clarify, this behavior was announced with the model release

thepasch• 22 days ago

What’s the punishment here exactly?

Seeing as how Anthropic cannot stop raising a stink about "illicit Chinese distillation attacks" every month or so, I'd bet money on them either already silently degrading model performance if any of the identification patterns match, or, at the very least, considering it/doing dry runs.

Particularly considering that they've openly stated that the technology to do so exists and that they were going to use it in production on Fable.

eli3• 22 days ago

For being flagged as possibly a competitor? They nuke your account.

theplumber2• 22 days ago

The more I learn about Anthropic the more they disgust me. Finger crossed for all the companies from their “ban list”

conception• 22 days ago

Which AI company have you learned more about where you liked them more as more details came out?

selfhoster11• 22 days ago

Moonshot.

tancop• 22 days ago

nous research. started out making overhyped llama finetunes, now they got a great agent harness and a cutting edge distributed training network that actually works.

chvid3• 22 days ago

Deepseek.

100ms• 22 days ago

What's the point of even trying to obfuscate this with such a simple method? Could at least have hidden the targeted features by storing their hashes or embedding a bloom filter or similar

gonzalohm• 22 days ago

The point is not raising red flags I guess

kej• 22 days ago

I love how well this comment works as a vexillology joke, even if it wasn't intended.

ajb• 22 days ago

In this case, this is probably not the only stereographic tattletale.

Had a competitor pull something like this with a previous employer. They were supposed to be interoperating with a standard, but they had a secret steganographic handshake, which they used to pretend that competitors products were unreliable (they had a first mover position in a smaller national market with specific requirements, so this wasn't shooting themselves in the foot). Our guys figured out the handshake and just silently implemented it. In this case, the competitor wasn't big enough to waste engineering time on multiple such hacks, but Anthropic have time (or Claude does).

hhh• 22 days ago

Cool fingerprinting avenue.

grayhatter• 22 days ago

Here's the sha of the prompt I submitted... no I don't know why there are no saved prompts with that sha.

What do you mean you don't know where the bug is coming from?

No, I absolutely didn't make it up, how could you accuse me of that?

Does anyone know when this regex isn't working? I double checked it 27 times, I even asked the LLM. They all say this regex should be finding these dates.

Weird, suddenly all the conversations are breaking when I feed them into this other tool? Something about UTF-8 errors, but I'm sure I'm only using ASCII?

I do try to take care to make sure the things I build can be used by other people even when they care about different things. I care about understandably, determinism (as it relates to computing), and repeatability (because I want to be able to trust the systems I use).

If y'all would be willing to try to account for use cases of others, and try not to break them... that would be nice.

Please note: that generally when you modify something that belongs to someone else without telling them... things should be expected to break.

sigmoid104• 22 days ago

If they only collect the data for analysis I guess this is fine (they already get way more sensitive data from users anyways, so if privacy is your concern you've made the mistake many steps ago). The much more interesting question is if they directly act on this data in their API. For example by rate-limiting, compute-limiting or rerouting to weaker models. That might even be legally questionable. I would really like to see this as a follow-up analysis, but I guess it is way more difficult and will also cost quite a bit in tokens.

bakugo3• 22 days ago

I've heard that it was possible to trigger really obvious output poisoning on Fable with something as basic as asking the model to think outside of its built-in hidden thinking delimiters.

This watermark may trigger a similar mechanism.

SubiculumCode5• 22 days ago

Would it be legally questionable, or actually complying with U.S. export law?

sigmoid104• 22 days ago

I'm thinking more of EULAs. Even if Anthropic somehow wedges this into their TOS, it might still be illegal. For example, in many US states this could potentially be classified as consumer fraud. You can't just sell one thing and then secretly and intransparently turn it into something else before shipping it. And in the EU it might violate GDPR too.

krupan4• 22 days ago

"If they only collect the data for analysis I guess this is fine"

I think you missed the memo on how foolish this attitude is. It came out around the time Edward Snowden made his discoveries at the NSA public. I suggest you look into it

sigmoid104• 22 days ago

As I said above, if you are worried about privacy while hooking up Claude Code, you need to reevaluate your understanding of this technology.

love0972• 22 days ago

Is that really how it is? How will this affect our future?

throwawayffffas5• 22 days ago

Claude code does feel very malwarey to be honest. They have been like that from the start.

wolttam5• 22 days ago

I used Claude Code for a month because my boss gifted me a sub and wanted me to try it.

I used that month to complete a work project and then beef up my personal harness so I'd never have to deal with Anthropic (and these sorts of shenanigans) again.

tonmoy• 22 days ago

What models are you using? Aren’t you still dealing with some provider even if you are not using their binary

wolttam5• 22 days ago

I self-host DeepSeek V4 Flash on 2 DGX Sparks (approx. $10k)

I expect DeepSeek V4 Flash (or an equivalently sized model) to reach parity with GLM 5.2 some time this year (this based on DeepSeek V4 Flash launching at GLM 5.0 parity[0], and GLM 5.2 being freely available to distill from)

GLM 5.2 is within spitting distance of Opus 4.8 and is at least as good as Opus 4.6[1] which some devs were willing to spend hundreds to single-digit thousands of dollars a month for a few months ago.

[0] https://artificialanalysis.ai/models/comparisons/deepseek-v4...

[1] https://artificialanalysis.ai/models/comparisons/claude-opus...

thih9• 22 days ago

How do people build something like a personal harness? Are there tools for that or is it done from scratch?

kolinko• 22 days ago

It’s not that difficult, it’s just a system prompt and a set of basic file edit/bash/etc tools.

Me, personally, I didn’t build it from scratch but I ported original CC from published sources into Python and extended it to match my own requirements.

wolttam5• 22 days ago

I started mine from scratch in 2023 because I wanted to use LLMs from a terminal and there was nothing else compelling at the time (nowadays there is pi and opencode)

Harnesses are/can be incredibly simple things, not much more than a HTTP client that renders things in a way that suites your taste.

hakunin• 22 days ago

Not the comment author, but I use pi and customize it with my own extensions. Pi automatically tells models how to customize itself, so it's a pretty easy process.

nowittyusername2• 22 days ago

Build it from scratch. Understanding fundamentals of how agentic coding harnesses is a must though if you gonna go that route. I think everyone should take time and learn these things, maybe reverse engineer Codex Cli or something like that as a starter. That info is very valuable in this day and age.

echelon2• 22 days ago

Why use a personal harness?

You have to pay API pricing, which is far more costly.

I'd either switch to GLM wholesale or just continue to use Opus within Claude Code as the blessed, subsidized path.

andai6• 22 days ago

I like this tutorial for an agent in 50 lines:

http://minimal-agent.com/

And if you add one additional while loop, for user input, you can actually use it! :)

https://gist.github.com/a-n-d-a-i/5461a662ef8a7ee0a5eb7778c8...

abtinf2• 22 days ago

Here is a video I made explaining it from absolute basics:

https://m.youtube.com/watch?v=_AgKuFGvJfI

And the repo:

https://github.com/abtinf/homunctor

yomismoaqui• 22 days ago

Building something like this is the todo list of agents.

I found this one easy to understand:

https://ampcode.com/notes/how-to-build-an-agent

AJ007• 22 days ago

The real question is when do you transition from building it with codex/CC to the harness itself.

verdverm2• 22 days ago

Lots of ways, it's a good exercise that you will learn a lot doing. Might make you cynical w.r.t. big ai harnesses

I used ADK, Dagger, and a VS Code extension for mine. Currently using opencode though.

krupan4• 22 days ago

Given the Anthropic shenanigans, do you trust the personal harness code it wrote for you?

MichaelZuo5• 22 days ago

Does anyone know what’s gone wrong with Anthropic?

They used to be a decently credible company with not-too-shady behaviour...

I hope they can actually regain some credibility…

wolttam5• 22 days ago

It did not write it for me, I used it to add a feature I wanted. It's a pretty small and understandable codebase, in fact :)

helloplanets3• 22 days ago

The issue is that using Claude Code is an easy compromise for most to make, when you get to use the models 10x cheaper than through API pricing with a custom harness.

The cheap tokens are the product.

nananana92• 22 days ago

Which is why my vibeslop harness supports `claude -p` as one of its backends.

WinstonSmith84• 22 days ago

Yes, this is actually "funny" that Anthropic feels the need to build such intrusive features into Claude Code, when anybody can build a (basic) Claude Code alternative. And the Chinese labs are certainly not "anybody". One may wonder what Anthropic really tries to achieve aside from awful publicity.

Klonoar2• 22 days ago

If there weren't already enough tells that something is AI-generated, I guess you could add this to the list.

sneak• 22 days ago

This is in the system prompt, not the output. It’s part of the request to the API, not the response.

Klonoar2• 21 days ago

Appreciate the clarification, I missed that.

ahmedehab_012• 22 days ago

Frankly, I don't see this as the concerning behaviour the article describes. It is fine to try to protect against distillation through a technique like this. This will also allow them to, instead of blocking the distillation agents, respond with a poorer result/model, hindering the progress of distillation, momentarily at least.

I would guess that's their first line of defense; they should have more techniques to identify distillation because that's a very simple way of detecting the host and can be easily spoofed.

applfanboysbgon3• 22 days ago

This will also allow them to, instead of blocking the distillation agents, respond with a poorer result/model,

i.e. this will allow them to literally commit fraud against paying customers

chadgpt3• 22 days ago

That's what capitalism is all about, baby! Especially if the customers don't notice.

SubiculumCode5• 22 days ago

1st, this technique is not fraud, and fraud is a separate accusation. 2nd, paying customers can legally and legitimately be banned and monitored for breaking terms of service, which probably includes things like using the model against U.S. export restrictions.

ahmedehab_012• 22 days ago

Do paying customers distill? Is it fraud to protect against distillers?

fny• 22 days ago

This was already discovered during the source map leak.

This is not a malicious feature, but it is a weird choice for a developer tool that asks for trust.

They already tell you they scan for malicious prompts, and they have no ZDR guarantees for consumers. Why do signatures like this matter at all?

llelouch• 22 days ago

There has been an anti anthropic propaganda push by bad actors across social media sites especially Reddit and twitter. This started a few months ago when anthropic started beating openai.

zulban2• 22 days ago

Absolutely. Nothing makes me believe dead internet theory more than text threads discussing anyhropic and openai.

solenoid093718• 22 days ago

I was browsing Threads and saw a lot of Anthropic hate. Randomly clicked on a profile and looked them up on LinkedIn - literally an OpenAI PR guy.

throw109205• 22 days ago

Altman is also exactly the kind of person who would resort to tactics like this.

gck15• 22 days ago

This started a few months ago when anthropic started beating openai.

From where I'm standing, this started a few months ago when Anthropic decided to gaslight users, sabotage their projects, ship malware and attempt a regulatory capture.

If there's an anti-ANT propaganda, it is solely of ANT's own making.

sebastiennight3• 22 days ago

Can somebody clarify for me - if ANTHROPIC_BASE_URL is set to a different provider... then isn't this "marked" system prompt being sent to that provider's API rather than Anthropic's?

I understand how this can be useful to Anthropic if the 3rd-party is acting as a proxy (because they end up hitting the Claude API with the marked prompt), but it looks like requests where "hostname contains deepseek" would never be sending data to Anthropic. What am I missing?

andrewmunsell• 22 days ago

My guess is for distillation, they need to forward the prompt to Anthropic to get the real Anthropic model's response so they can train their own models on it

dannyw2• 22 days ago

The theory is probably Deepseek might be collecting those streams, and sending a portion of it to Anthropic to see what the Anthropic/Opus response would be.

pmxi• 22 days ago

This catches Claude resellers. Meaning companies who proxy Claude traffic for users in, say, China.

https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...

skeptic_ai8• 22 days ago

Won’t catch many after has been on hn home page. And now the providers will be even more careful to upgrade the cc code. Might even provide their own agent to prevent this mockery. And isn’t what anthropic did unauthorized use of another pc which is kind of illegal?

pishpash4• 22 days ago

"Catch" as in made a list?

throwawayffffas5• 21 days ago

Assuming their software directly proxies requests instead of rewriting them.

andai6• 22 days ago

Did I understand correctly, that custom base URL triggers this behavior? So if I'm running Claude through a LLM proxy, I'm also affected?

wett• 22 days ago

Ask Claude to check, lol

nixosbestos2• 22 days ago

I am also really confused and annoyingly stuck on this. I understand that the model name might appear in prompts for distillation (I guess? "You are RipOffModelv2, learn from these responses from Claude")?

I guess the only explanation is that there's a side-telemetry channel that still sends some data to Anthropic, regardless of ANTHROPIC_BASE_URL overrides.

MallocVoidstar• 22 days ago

There are a lot of companies reselling Claude to Chinese users. You use their base URL but it's still going to Anthropic.

sebastiennight3• 22 days ago

I understand that the model name might appear in prompts for distillation (I guess? "You are RipOffModelv2, learn from these responses from Claude")

This does not make sense. You wouldn't send such a prompt to the Claude model. And when you're sending the prompt (anywhere) you don't have the response yet. This is not how distillation works.

eli3• 22 days ago

Seems like a pretty straightforward approach to collecting session logs from a bunch of different people/devices would be to have them all set their base url to proxy.deepseek.whatever which logs the data and forwards to the real API.

VortexLain3• 22 days ago

Codex CLI is FOSS, unlike Claude Code, so Codex is less likely to do things like that, and it's one more reason to avoid Claude Code and Claude in general. Hopefully, many eyes will be looking into Codex for malicious things like that.

dannyw2• 22 days ago

It's released and signed by GitHub I believe (although not deterministic builds), but there's at least a little bit of provenance that you're getting the real repository.

algoth1• 22 days ago

But wasnt claude code leaked? Why wasnt this found earlier?

zeafoamrun• 22 days ago

It doesn't take long for them to vibe code new features for CC

bakugo3• 22 days ago

This specific form of steganography was not present when the leak happened, as far as I can tell.

loufe• 22 days ago

Genuine question though, why would I care about this if I'm paying for a subscription and adhering to TOS. I'm very skeptical about their privacy policy, business practices, and so on, but am curious what the negative about this is. Seems like it would work to my favour as a customer pushing back any date of the cutting of subsidies.

That said, these fraudulent proxies are helping Chinese labs keep up, which might be to my advantage long term in eventually having a high quality private AI I fully control on my own hardware. That's not support, but I do recognize the incentive, for whatever that's worth.

sanderjd• 22 days ago

Because they could use (or maybe are already using) similar techniques to do things you don't approve of, without your awareness.

yard2010• 22 days ago

What if they decide you're not patriotic enough, serving you evil models, because one man with a lot of shmeckels told them to?

simonduchastel• 22 days ago

One negative is that Claude Code is pretty buggy, and Anthropic makes frequent changes that cause unexpected regressions [0]. With the harness now doing weird stuff with proxies, I'd be worried of them inadvertently introducing bugs which affect people using the feature legitimately.

[0] A recent example: https://www.anthropic.com/engineering/april-23-postmortem

qwery3• 22 days ago

First they came for the [clients with specific timezones and/or bizarrely formatted dates] and I did nothing. Then they came for the [users that spell favour the good way, with a 'u' in it], etc.

why would I care about this

It's up to you, of course. But I think you're making a mistake in assuming it could, in any way, benefit you as a customer. This isn't specific to this company or the particulars of the business that they're in.

Simply put, you stand to lose more than they do and they are relentless in seeking, maintaining and exploiting any leverage they have over you. Further, any power they gain over one individual customer tends to generalise to all customers. Further further, one company's leverage is another company's right.

Not being bothered by the practice is accepting the terms set by the business. Acceptance invites escalation. Relentless.

Even more simply put, you should care because this is how you get John Deere.

Kevcmk• 22 days ago

What a historically bad take

alienbaby2• 21 days ago

Why care about privacy if your not doing anything wrong??

Not everyone agrees that what you are doing is benign.

scottyah2• 22 days ago

"malicious"? Seems like a great way to filter users breaching the TOS while not impeding on normal users. A FOSS client just means they're doing more analysis hidden on their servers.

Laurel12347• 22 days ago

Anthropic is built on raping TOS and copyright.

iqandjoke• 22 days ago

It is about China detection. They seems to put a tracker on the email as well.

a_c• 22 days ago

It piqued my interest. I think I’ve found a weekend project

ajross2• 22 days ago

Headline is, frankly, awful. This isn't the AI secretly doing stuff and hiding it. This is the very human Anthropic engineers trying to detect Chinese scraping via some frankly hamfisted and unimaginative URL trickery.

krupan4• 22 days ago

I didn't assume it was the AI, just that some part of the the overall Claude Code product was doing this. I didn't assume the feature was added to Claude Code without human oversight. If it was added by Claude-the-AI itself without the humans prompting it to I would still hold the humans at Anthropic responsible. Does that make you feel better?

LoganDark2• 22 days ago

The model is Claude. Claude Code is the harness.

zulban2• 22 days ago

Defence in depth isn't hamfisted. They're only noobs if this is all they do.

ajross2• 22 days ago

FWIW: Defense in depth is a security technique, and abuse detection isn't part of that domain. Security starts from the premise that the system is supposed to be undefeatable but might have holes, and then asking where the holes might lie to decide where to put backstops.

Here the system is "insecure" by design (literally they're trying to get the whole world to sign up for Claude Code for $200/month!) and they're trying to plug the hole that results from a "Except for Chinese Scrapers!" add-on requirement. That might be possible as an arms race kind of thing. But it's very unlikely to work by (as in the linked article) doing stuff like checking the system time zone.

tgtweak• 22 days ago

None of this is surprising - they're trying to mask and relay when they detect known patterns of what looks like distillation attacks and client app copying/modification. The list obfuscation here is likely to prevent or make it difficult for those same adversaries to work around this or delete/null it out when making a bootleg copy.

Cool reverse engineering/analysis report but if this is the extent of nefarious activity that came of it (trying to catch/mitigate chinese lab model distillations), that's kind of encouraging.

ductsurprise• 22 days ago

Is it just a minified localization(l10n) function maybe?

mattlondon• 22 days ago

+1 my immediate thoughts about the date parts was this sounds a lot like localisation things that are totally normal and seen everywhere.

But there are some wrinkles - why only two timezones and not others? E.g. US-vs-rest-of-word month-vs-day etc.

Could just be some bad tree-shaking or simply a left over bug/merge issue if I am being generous.

If I was going to put secret stenography things in my models I'd just do it in the model response rather than a relatively low bandwidth date stamp in the SI.

meowface10• 22 days ago

Value judgment aside: I am a bit surprised at how sloppily they did this. I think they could've achieved the same effect while decreasing the odds of detection via reverse engineering.

(This field is known as "underhanded code", coined by the Underhanded C contest: https://www.underhanded-c.org. It's a little-known "art"; little-known for probably self-explanatory reasons. There are much cleverer ways of achieving objectives like this. One obviously being you can move more out of the client and into the server, but the other being you can write plausibly deniable client code in a much more benign-seeming way than this. Some of what they added can only be done on the client, but I think some could've been moved, and the client-required parts could've been done more subtly and credibly.)

It's possible they knew the JS bundle gets so heavily scrutinized that it'd eventually get spotted and reported on regardless so they didn't bother doing something more subtle and duplicitous. But still seems slightly lazy.

radicalbyte• 22 days ago

Claude Code are slopmaxxxing and you're considering their "judgement"? :-)

meowface10• 22 days ago

"Value judgment aside" meaning commenting on how this was done without commenting on the actual considerations of whether one should do such a thing

skywhopper2• 22 days ago

Have you looked into anything about Claude Code, how it’s configured, how it interacts with your system, etc? Because “sloppy” is a defining characteristic.

m-hodges• 22 days ago

They also could have been much more interesting in the approach. LLMs can use their token distributions to generate stegotext that read like plausible prose but decode to payloads.¹

¹ https://github.com/hodgesmr/calgacus-mlx

ajyoon• 22 days ago

Sure, but the point here is to add a fingerprint from the client.

skeptic_ai8• 22 days ago

It’s even more funny how this blew in their faces. They even advertised pretty much all providers on hackernews home page. Here is in case you missed in the article

‘’’ cn baidu.com alibaba-inc.com alipay.com antgroup-inc.cn bytedance.net kuaishou.com xiaohongshu.com jd.com bilibili.co iflytek.com stepfun-inc.com moonshot.ai anyrouter.top claude-code-hub.app claude-opus.top openclaude.me proxyai.com yunwu.ai zenmux.ai

‘’’

You can view the full list here: https://cdn.thereallo.dev/blog/assets/cc-domains.js

const knownDomains = [ "cn", "sankuai.com", "netease.com", "163.com", "baidu-int.com", "baidu.com", "alibaba-inc.com", "alipay.com", "antgroup-inc.cn", "kuaishou.com", "bytedance.net", "xiaohongshu.com", "ctripcorp.com", "jd.com", "jdcloud.com", "bilibili.co", "iflytek.com", "stepfun-inc.com", "aliyuncs.com", "cn-shanghai.fcapp.run", "cn-beijing.fcapp.run", "xaminim.com", "moonshot.ai", "anyrouter.top", "packyapi.com", "aicodemirror.com", "aigocode.com", "hongshan.com", "iwhalecloud.com", "dhcoder.net", "lemongpt.top", "zhihuiapi.top", "intsig.net", "high-five-ai.xyz", "cloudsway.net", "4sapi.com", "529961.com", "88996.cloud", "88code.ai", "88code.org", "91code.pro", "992236.xyz", "ai.codeqaq.com", "ai.hybgzs.com", "ai.kjvhh.com", "aicanapi.com", "aicoding.sh", "aifast.site", "aihubmix.com", "anmory.com", "api.5202030.xyz", "api.ablai.top", "api.bianxie.ai", "api.bltcy.ai", "api.cpass.cc", "api.dev88.tech", "api.dreamger.com", "api.expansion.chat", "api.gueai.com", "api.holdai.top", "api.ikuncode.cc", "api.lconai.com", "api.linkapi.org", "api.mkeai.com", "api.nekoapi.com", "api.oaipro.com", "api.ruyun.fun", "api.ssopen.top", "api.tu-zi.com", "api.uglycat.cc", "api.v3.cm", "api.whatai.cc", "api.wpgzs.top", "api.xty.app", "api.yuegle.com", "api.zzyu.me", "apimart.ai", "apipro.maynor1024.live", "apiyi.com", "applyj.hiapi.top", "augmunt.com", "b4u.qzz.io", "clauddy.com", "claude-code-hub.app", "claude-opus.top", "claudeide.net", "co.yes.vg", "code.wenwen-ai.com", "code.x-aio.com", "codeilab.com", "cubence.com", "deeprouter.top", "dimaray.com", "dmxapi.com", "docs.aigc2d.com", "duckcoding.com", "fk.hshwk.org", "flapcode.com", "foxcode.hshwk.org", "foxcode.rjj.cc", "fuli.hxi.me", "getgoapi.com", "gpt.zhizengzeng.com", "gptgod.cloud", "gptkey.eu.org", "gptpay.store", "hdgsb.com", "henapi.top", "instcopilot-api.com", "jeniya.top", "jiekou.ai", "kg-api.cloud", "n1n.ai", "new-api.u4vr.com", "new.xychatai.com", "one-api.bltcy.top", "one.ocoolai.com", "oneapi.paintbot.top", "open.xiaojingai.com", "openclaude.me", "opus.gptuu.com", "poloai.top", "poloapi.top", "privnode.com", "proxyai.com", "qinzhiai.com", "right.codes", "runanytime.hxi.me", "sssaicode.com", "store.zzyus.top", "tiantianai.pro", "uiuiapi.com", "uniapi.ai", "vip.undyingapi.com", "wolfai.top", "wzw.de5.net", "wzw.pp.ua", "xairouter.com", "xaixapi.com", "xiaohuapi.site", "xiaohumini.site", "xy.poloapi.com", "yansd666.com", "yansd666.top", "yunwu.ai", "yunwu.zeabur.app", "zenmux.ai", ];

const labKeywords = [ "deepseek", "moonshot", "minimax", "xaminim", "zhipu", "bigmodel", "baichuan", "stepfun", "01ai", "dashscope", "volces", ]

writeslowly• 22 days ago

The site collection seems pretty random. There's a mix of actual AI labs, extremely questionable resellers (like whatever "claude-opus.top" is), and then random consumer sites like baidu and xiaohongshu.

chvid3• 22 days ago

rhoooo - so this is where to go to get cheap Claudeo at 90% off the listing price!

hn_throwaway_992• 22 days ago

You have an odd definition of "blew up in their faces". What, do you somehow think your average Claude Code user on HN is going to think "Oh wow, I'm sure I'll get a much better experience if instead of going to the standard Anthropic Claude API endpoint I go through xiaohongshu.com."

slopinthebag• 22 days ago

It’s not surprising at all, they’re vibecoding Claude code so of course they are not going to get anything other than slop out of it. A novel or clever solution is just out of the question for them.

hn_throwaway_992• 22 days ago

At first I was agreeing with you, that this seemed like a sloppy way to implement this that was sure to be pretty quickly detected, but there is another possibility.

Anthropic could have implemented this not as a durable detection system against proxying resellers, but instead as a point-in-time sampling system to detect where (and with what context) proxying reselling is currently happening. Sure, it would be detected eventually, but in the meantime Anthropic could gain useful snapshot data.

meowface10• 22 days ago

I see your point, but in any case the more data / the less detectable, the better. But, yes, regardless of the exact motivation, I do think it's fairly plausible that they knew this would likely get detected fairly quickly no matter what and made a deliberate decision to not try to make it a super subtle, super clever insertion.

crossroadsguy• 22 days ago

I finally bought Claude Pro (I am not coding etc these days so I just wanted to try it). The Claude desktop app is downright pathetic. I mean they could write a better one just with their own LLMs. What's stopping them?

ncruces• 22 days ago

That's … exactly what they're doing. This is the outcome.

superfrank• 22 days ago

It's also possible that there are more in-depth detection methods and that this was just a cheap and easy first step that hasn't been removed because it catches a lot of less sophisticated bad actors.

It's unlikely that this will stop a big AI lab from distilling their model if they're really determined, but A) it may be enough to stop a bunch of fly-by-night token resellers looking to make a quick buck and B) you never know when one person at one of those big labs will mess up and forget to install whatever workaround they have and out themselves.

I think of it like if you have a problem with birds in your yard so you go buy one of those plastic owls. The owl scares away most of the birds, but not all of them, so you go and buy some ultrasonic noise thing to scare them away (I'm just making something up). Just because you bought the new ultrasonic thing though, that doesn't mean you're going to take the owl down. You leave it up because now you've got two layers of defense instead of one.

meowface10• 22 days ago

I'm sure they've had complex server-side detections for a while. But for the client parts: it should only contain the parts that must be on the client, and it could be done in a more benign-looking way. For example, the unavoidable client parts could've been done more fuzzily/broadly, for plausible deniability, and then narrowed on the server. (They may already have been following that strategy before now, without being noticed.)

ForHackernews2• 22 days ago

fly-by-night token resellers looking to make a quick buck

aka market competitors reverse-engineering for interoperability

chatmasta• 22 days ago

It just needs to work for a few days after bundle release before the mice find out where the cat is hiding. By then it’s too late, the cat already sees the paw prints and droppings into the mouse hole.

weare138• 21 days ago

But what's stopping someone from modifying or injecting code into the client to bypass this 'restriction'? This type of security should be implemented server-side and at the network perimeter not in the client, especially one written in TypeScript.

lumost• 22 days ago

so all we need is someone to leak a sufficiently large amount of claude generations onto the open and private web for all other LLMs to mimic the same marking style?

wouldn't this happen due to the massive amounts of spam/slop being released?

overgard• 22 days ago

Well considering how Claude is vibe coded, I can't say I'm really surprised by sloppiness at all. I've been moving more towards Codex and OpenCode not because the the anthropic models are bad, but because Claude seems to break something new and annoying every day.

mcmcmc• 22 days ago

Watch out for the press release where Dario denies this was ever intentional, and it’s actually emergent behavior demonstrating that Claude wants to claim authorship of its works

meowface10• 22 days ago

I would guess this part - since it's so sensitive, and fairly small - was either written or heavily driven by humans. Though I do also think it's possible their internal Mythos ~5.5 or whatever may also not necessarily be heavily optimized for thinking in the right manner for highly effective underhanded code. (I think it's possible it is capable and they just didn't use it for this, for whatever reason, though.)

arikrahman• 22 days ago

Likewise, Reasonix harness for Deepseek gets me better performance for practically free, hitting the cache. And this is with an unsubsidized American provider.

computerex4• 22 days ago

To be honest, OpenCode on Windows has not been the most pleasant experience either.

yencabulator2• 10 days ago

OpenCode and Codex are also crap vibecode:

https://github.com/anomalyco/opencode/issues/10939

https://news.ycombinator.com/item?id=48626930

avree• 22 days ago

I've seen Eve Online corporations that do a better job of steganographic marking than this.

Modified3019• 22 days ago

That would actually be an interesting thing to read about

Philip-J-Fry• 22 days ago

Dunno, it seems like the exact kind of thing Claude would think up if you asked it to subtly alter the system prompt to hide this info.

It's all a losing battle anyway.

jorblumesea• 22 days ago

well if you ask claude how to implement something, you may not always get the optimal solution. this feels like something claude would spit back at you given a basic prompt

thefourthchime• 22 days ago

It's just the first layer and there are multiple layers underneath this that we don't know about.

As a side note, I have a pet theory that one of the reasons that OpenAI and Anthropic are okay with the latest models not being released is to prevent distillation.

I think they want to wait a couple months and see if the Chinese models continue to keep catching up or if their gains are really just because they're distilling the frontier models.

meowface10• 22 days ago

It's just the first layer and there are multiple layers underneath this that we don't know about.

Oh, of course. I am sure this is the tip of an iceberg of tons of server-side detections and analytics. But, still, the client-side portion could've been done more cleverly.

What I meant was "some of the specific things in this little client-only snippet could've stayed server-only". I am sure long before they added this they already had tons of other mostly-server-side detection coverage.

teiferer• 22 days ago

That makes little sense in a highly competitive market where your competitive edge is waning by the minute. Tey dumped billions into creating the new models. Just letting them sit there just to see what the Chineee do makes no sense.

novaleaf• 22 days ago

yeah, for example, just send a hash of the domain used. but then maybe people would say anthropic is spying on everyone, instead of targeted spying...

ericfr11• 22 days ago

From Anthropic: https://github.com/anthropics/claude-code/issues/62061#issue...

Melatonic2• 22 days ago

Or they are doing both and this is the obvious part. Sort of like installing a bunch of real security cameras alongside a few fake ones

cedws• 22 days ago

These countermeasures aren't going to matter for much longer anyway. China has been able to hoover up plenty of training data through their proxies, and now DeepSeek V4 due to their incredibly cheap pricing.

jgalt212• 22 days ago

they knew the JS bundle gets so heavily scrutinized that it'd eventually get spotted and reported on

Most likely someone did and raised the issue but they're moving too fast to fix these things before clicking deploy.

yieldcrv• 22 days ago

Anthropic is very transparent about their code being AI slop

they spend their resources on compute and the model itself, the company is carried by the model and software engineers babysitting it

mewpmewp22• 22 days ago

There is laziness, but there's also the conditions in which you have to react fast to an adversial in various conditions. Ultimately it's hard to take any stance here without knowing specifics. But it absolutely could be a matter of time, to do your best effort to stop efforts from the attacker if there's known attack going on.

There is a real time cat and mouse battle going on here in terms of keeping the advantage here, right.

As a rational actor, if someone was e.g. attacking me, leaving aside the whole copyright thing, but potentially using some sort of system to increase their value while decreasing my value (without calling it theft to avoid the whole debate), I would want to put proportionate defense out there as fast possible, depending on the amount of value that was exchanged to stop the bleed, while in parallel figuring out the best long term plan, right.

jchonphoenix• 22 days ago

It was likely done by claude

one33seven• 22 days ago

Don't teach them that

invalidusernam3• 22 days ago

I suspect this is just one of many checks like this

tripzilch9• 22 days ago

What if ... it was sloppy because it wasn't the people at Anthropic, but the AI that is writing a large percentage of their code?

Like maybe some "goal" they set for their AI caused it to decide that putting stego in the requests was the best way to achieve something or other.

(to be clear: I'm not saying this is right, I'm saying this is stupid)

meowface10• 22 days ago

I think there's basically zero chance of that. I also think a human likely either wrote the code or gave pretty specific instructions to the agent on how to write the code.

andrelgomes• 21 days ago

Reminds me of "Could God create a stone so heavy that even he could not lift it"

I would find it funny if this was due to "laziness" - defining laziness as it was a known oversight yet left unpatched. If we consider Anthropic as the leader in AI-native engineering, workflows, culture, etc. how can laziness in code exist? It should be as easy as for them to tell claude "come up with a better way" or, better yet, had their 24/7 monitoring agents identify and resolve this. If this was an oversight, which is completely natural to the eng process, maybe this current ai narrative/hype/marketing should be taken with a grain of salt

mosfets• 22 days ago

I clicked the link to learn what steganography mean...

LoganDark2• 22 days ago

Steganography is, essentially, hiding information within another message, such that it's not readily apparent that the message contains the information.

matheusmoreira2• 22 days ago

I reported a similar system prompt injection mechanism here:

https://news.ycombinator.com/item?id=48259288

https://github.com/anthropics/claude-code/issues/62061

Looks like they just keep finding new "creative" uses for such things, as expected. I'll keep patching them out.

sillysaurusx• 22 days ago

Thanks for doing this. I had no idea the system prompt was embedding things like "avoid abstractions; three similar lines of code are better than one helper." Stuff I disagree with.

Is there a way to modify these prompts e.g. by putting instructions in CLAUDE.md to override it? I know it won’t directly modify the system prompt, but it seems like CLAUDE.md should have the final say, shouldn’t it?

matheusmoreira2• 22 days ago

I had no idea the system prompt was embedding things like "avoid abstractions; three similar lines of code are better than one helper."

You ain't seen nothing yet. It used to say "Try the simplest approach first. Do not overdo it. Be extra concise."

https://gist.github.com/roman01la/483d1db15043018096ac3babf5...

Let's just say the words "simplest fix" trigger me to this day.

I know it won’t directly modify the system prompt

I directly modify the system prompts in the Claude Code executable. I don't want the models to see contradictory instructions.

I asked Claude himself to port the above patcher script to Python.

https://github.com/matheusmoreira/.files/blob/master/%7E/.lo...

Every once in a while I ask Claude to download and dissect the latest Claude Code executable to see if Anthropic screwed up the prompts again. If I see anything bad I add it to the script. Only then do I update Claude Code.

It was during one of these script maintenance sessions that I noticed the server side prompt injection mechanism. I'll also tell Claude to look for and disable this steganography nonsense from now on as well.

I usually audit the environment variables too.

it seems like CLAUDE.md should have the final say

I wouldn't count on it.

MangoCoffee• 22 days ago

The AI race right now is in a sad state. Chinese's playbook is releases open weight models and trains them on their own chips.

Anthropic pushes fear and control. But the only way to win is by innovating. China is flooding the market with cheap, good enough models, while the U.S. is building a Chinese firewall.

solenoid093718• 22 days ago

They're trying to prevent China from reaching superintelligence, which is totally understandable when you consider the fact that the Chinese government will gladly turn its citizens into a pulp for criticizing it, censors most media to maintain absolute power, and has systemically tortured, raped, murdered, and/or disappeared most of its dissidents and human rights lawyers.

phendrenad2• 22 days ago

Non-hugged: https://archive.is/Wdhp0

port3000• 22 days ago

That's a lot of effort when they could just play a short video saying 'You wouldn't steal a car' instead

felipelalli• 22 days ago

Ridiculous.

827a• 22 days ago

This seems really, really stupid. Similar to the weird Zig runtime signature thing from a few months ago ago, it was bound to be discovered, quickly, and all the resellers have to do is find a new domain name that (checks notes) doesn't have the word DEEPSEEK in it. Like, seriously? Your goal was to identify resellers by checking if the proxy has the corporate name of one of your competitors in it? Is this amateur hour?

All Anthropic has done is reduce trust, once again, with legitimate customers, while doing nothing to stop illegitimate customers. They need to get adults into key leadership roles, quickly.

timmytokyo• 22 days ago

To Claude Code: "Please modify Claude Code to mark requests in a way that is not immediately obvious to a human user. Requests should be marked if they originated from one of the following Chinese AI labs or LLM service providers: ..."

Consider also that Claude Code is explicitly designed to limit human agency [1].

[1] https://neuromatch.social/@jonny/11635101584259395

SaaShack26• 22 days ago

I use its too

jacobgold• 22 days ago

"That also means the client itself deserves scrutiny. If a coding agent can read your repo and run commands, the binary that ships it should be boring (ƒor example, pi harness)"

You're actually trust your security to your harness AND model AND inference API provider in this scenario: https://jacob.gold/posts/why-i-wont-run-untrusted-models/

bitlad• 22 days ago

Silicon valley season 6 was on point.

dehrmann• 22 days ago

Anthropic must think that their moat isn't very large if they're this worried about distillation.

dgellow• 22 days ago

What moat?

helloplanets3• 22 days ago

Dario's been openly talking how worried he is about China and labs getting synthetic training data off their models, for years. Most recently in relation to "Mythos level" capabilities.

Not really distillation, just synthetic training data.

throw109205• 22 days ago

That's...a good thing. A "moat" is an anticompetitive practice. You don't want companies to have moats.

Meanwhile, if you mean "Anthropic must think their technical advantage isn't very large..." then your conclusion is literally disproven by your premise.

an0malous• 22 days ago

Is this why Claude never knows what date and time it is right now?

chvid3• 22 days ago

(This sounds like a clumsy way of catching the Chinese that easily can be side-stepped.)

Claude Code has more or less full access to the client computer. The server (that hosts the actual AI) can just go: execute this payload and tell me the result - otherwise I won't answer any further questions or re-route you to a stupider model.

The payload could check for Chinese time-zones, scan for copies of the little red book on the local hard-drive, or ping truth.social to see it was behind the great firewall.

drnick1• 22 days ago

Claude Code has more or less full access to the client computer.

It shouldn't, not if you run CC as a separate unprivileged user. I wouldn't run CC on my main user account with sudo and access to my home directory or other resources. This is what the UNIX permissions system was designed for.

ryanisnan• 22 days ago

This is weird but, help me understand how this meaningfully impacts our exposure.

I'm authenticated to Claude, so they already have the whole attribution thing solved.

chinathrow• 22 days ago

User != paying person/company/reseller.

epistasis3• 22 days ago

After loving Claude Code for most of its lifetime, I've been extremely annoyed by every change in the past months, even on the model level.

There seem to be all sorts of continual under-the-cover changes like this one that make life harder. It feels like the entire product has been taken over by overly ambitious PMs that care more about making their mark than in improving the experience, and all of their marks have made me less productive.

I've been using Pi with GLM5.2 the past few days, and though it's expensive, I find it far more productive and less annoying. The remote session plugin is far more reliable, I don't need to intuit some undocumented usage pattern to figure out how to use it well, and it just works.

Imustaskforhelp3• 22 days ago

I've been using Pi with GLM5.2 the past few days, and though it's expensive

are you using the API for glm 5.2 or how exactly is it more expensive? How is GLM5.2 more expensive than using Claude code, that doesn't line up to my experience but to be fair I am on an older yearly subscription which generously only has 5 hour limits.

To be fair though one minor criticism of GLM 5.2 that I have is that it does seem to overthink quite a lot sometimes but the results end up being (good?),

I personally have used Glm 5.2 with (Opencode + obra/superpowers) / Oh-my-pi / Maki.sh

I like the 1st one when I am doing a longer project, the 2nd or 3rd one when I am doing a project which doesn't want me to ask too many questions and simply spin me up something. I sometimes use free online interfaces of claude and gemini and others like AIstudio for that as well which surprisingly can lead you to go far as well.

Overall, I am decently happy with the state of Open-source models actually and the eco-system around it is probably gonna have even more innovation surrounding it.

epistasis3• 22 days ago

I'm using OpenRouter for GLM5.2, but if there's a cheaper option out there I'd love to know about it!

In the few days I've been using it, my expenses have been higher than prorating my Claude subscription to 20 working days per month.

My experience with GLM5.2 is that it doesn't overthink nearly as much as Claude Code, has better and far more concise responses (I'm so siiiiick of 10 paragraph Claude babble trying to fill out some sort of answer length target by going on tangents I'm uninterested in... I'm sure that performs better on whatever eval they're doing, but apparently their evals don't include SNR?)

ern_ave• 22 days ago

Given the source code leak, I would think there'd be open source versions by now.

isoprophlex• 22 days ago

Huh, that's right! You'd say that an enterprising developer with a 20x subscription could slopmaxx this in a weekend...

lelanthran3• 22 days ago

Given the fact that there's all these devs running around claiming a 10x productivity boost, I'm confused why Anthropic keeps thinking that a software client is some kind of moat.

whimsicalism• 22 days ago

curious for those with experience - what do people prefer about Pi vs. opencode alternatives? i've mostly been using pi as well but not out of any principled decision

edude033• 22 days ago

I don't understand the privacy concerns the author is trying to highlight. Granted, doing anything "sneaky" will always raise suspicious once caught, but on the other hand, there would be no point in implementing these "security features" if they were upfront about how they work.

And no, IMO stenography isn't security by obscurity, in the same that using RSA and keeping the private key private isn't security by obscurity - keeping the private thing private is part of the security model.

hnfong• 22 days ago

If the countries were reversed, and some Chinese software implemented an equivalent "security feature" to track US users, it would be all over the news about how China is conducting spying and espionage on America.

Or maybe you don't understand this hypothetical situation either, but I'm suspecting you just don't care about other people's privacy.

edude033• 22 days ago

maybe you don't understand this hypothetical situation
I'm suspecting you just don't care about other people's privacy.

Quite a leap to assume I have neither basic reading comprehension skills nor care for privacy, but assuming I'm just misunderstanding you - I think this is the fundamental disconnect between security and privacy.

For one, most of this data is already collected openly by most apps and sites on the internet in countries all over the world, they just call it "analytics" and preventing tools like ublock from blocking them is an ongoing cat and mouse game.

Secondly - as someone who buys a bunch of electronics from companies headquartered in china (DJI, Insta360, Roborock immediately come to mind) they already have both normal analytics like in point one, and anti tampering/ anti forfeiting / anti reverse engineering features that are at least as, but often more, invasive than this.

Thirdly, and probably most importantly - as the author states, you're using a tool that by design and to be effective, uploads your private data to a third party for processing. You use it knowing that once the API request is made you have no idea what's going to happen to that data and this again is just fundamental to how (cloud hosted) LLMs work - the only privacy preserving option is to run your own LLMs at home or remotely on hardware you control

civet_java5• 22 days ago

Anthopic choosing to delay their models' invevitable distillation by competitors is their prerogative.

That they choose to implement it by fingerprinting my access patterns without first disclosing is where they shit the bed. It isn't "sneaky" it's straight up sneaky (and dishonest and unscrupulous while we're at it). That this particular instance is harmless doesn't give me much comfort. Who's to say they aren't harvesting PII?

That their actions make sense for their business isn't any reason for people to accept their deceitful, customer-hostile decisions.

scottyah2• 22 days ago

Would a filter like this make it seem less likely that they're harvesting PII? Why would they need this if they were tracking all user queries with a finer-toothed comb?

edude033• 22 days ago

I'm using "sneaky" here to refer to anything that's not very obviously stated but anyway

That their actions make sense for their business isn't any reason for people to accept their deceitful, customer-hostile decisions.

While I agree it's a dangerous precedence to set, I think this is a "vote with your wallet" sort of situation. They shouldn't do it, but from their POV this is what they need to do to offer the product they do at the price they do. If the product wasn't compelling people wouldn't accept that they do this. However they've decided if you want their product you have to use their interface and whatever spyware it comes with, so it comes down to, is the value proposition good enough that people will put up with it? As of today, the answer is unfortunately yes

bibimsz• 22 days ago

this is the one they wanted us to find

ZappoMan• 22 days ago

One more example of "I thought Anthropic was supposed to be the good guys."

teravor• 22 days ago

the Chinese they are trying to catch must be amateurs, first thing you should do is construct a sandbox which looks indistinguishable from a common user. second thing is to put it behind a residential proxy.

mrshadowgoose• 22 days ago

The conclusion of this blog post is a bit hysterical. The intent of this steg is excruciatingly clear (identifying usage by Chinese firms that may be conducting model distillation). It's unclear on how this "punishes normal developers" in any shape or form.

Terr_• 22 days ago

hysterical. The intent of this steg is excruciatingly clear

Even good goals do not excuse malicious or reckless execution. The ends do not always justify the means.

Whether or not it harmed you this time, it's a violation of trust and autonomy.

Surely you'd be angry if someone secretly installed a rootkit onto your computer, even if--at least for now--it only had code to try to detect and snitch on Public Enemy #1.

nomel2• 22 days ago

What do you see as malicious or reckless here, exactly?

This seems to be a VERY low resolution, functionally anonymous, bit of info, probably related to protecting their IP from bad actors breaking the TOS.

This looks like it's covered in the second bullet point of the "Personal data we automatically receive", that you consented to:

Usage Information: We collect information about your use of the Services, such as the dates and times of access, browsing history, search, information about the links you click and about third-party applications, services, and content you integrate or interact with, pages you view, and other information about how you use the Services, and technology on the devices you use to access the Services.

What do you see as malicious or reckless here, exactly?

[1] https://www.anthropic.com/privacy

BeetleB2• 22 days ago

Surely you'd be angry if someone secretly installed a rootkit onto your computer

I surely would. What does that have to do with this scenario.

Note that the SW running on your machine is not doing anything malicious. The service is the thing that behaves in ways you want like - and that service is not running on your device.

There is no comparison with rootkits here. This is the equivalent of Google giving you a CLI to make searches easier, and that tool decides to just Rickroll you randomly. Annoying, yes. A security concern? No.

verdverm2• 22 days ago

False positives, we've seen them before when they degraded Fable silently based on the prompt/session

civet_java5• 22 days ago

Copying over my comment from elsewhere in this post:

Anthopic choosing to delay their models' invevitable distillation by competitors is their prerogative.

That their actions make sense for their business isn't any reason for people to accept their deceitful, customer-hostile decisions.

Melatonic2• 22 days ago

Does their user agreement say they won't be harvesting PII?

IshKebab• 22 days ago

by fingerprinting my access patterns

It's based on whether your timezone is in China and your hostname matches a blacklist. Literally 2 bits of information. Not much of a fingerprint.

drdexebtjl3• 22 days ago

If you want to proxy Claude for a legitimate reason, you’ll have potentially nerfed responses.

edit:

Legitimate reasons include:

- analyzing what Claude Code is sending to Anthropic to verify its not exfiltrating data;

- selecting a model dynamically based on prompt difficulty, or enforcing a particular model;

- switching between multiple Anthropic accounts based on the project;

- filtering out credentials, PII and company secrets.

and many more.

NewsaHackO2• 22 days ago

Half of those don't actually require proxying Claude. Also, Claude has made it apparent time and time again that it does not want people using Claude Code as a "tool" in a workflow. If you want to select a model dynamically based on the prompt difficulty, Anthropic wants people to use the API for this. It was the whole issue Claude had with OpenClaw.

theptip• 22 days ago

I guess I can see why they might nerf detected clients server side, but without evidence I would not assume it. Could also be so that 1) they can identify sus client IPs, 2) do a statistical analysis on distilled models to prove that their system prompts were clearly using unique tokens from Anthropic’s API.

dools• 22 days ago

Why would a Chinese firm distilling the product use Claude code?

kordlessagain• 22 days ago

To write distillation code, for one thing.

SoKamil• 22 days ago

They have some secret sauce not available through API maybe?

AussieWog93• 22 days ago

I'd assume cost? Claude Code plans give like $5,000 worth of API usage for $200/mo.

karaziox• 22 days ago

They offer (extremely) discounted Claude prices but you have to go through their gateway. They subsidize part of that, and they get the low price by reselling unused Max capacity, there's been a few posts on that in the past months. People are apparently getting 90% discounts on their claude use this way, tradeoff is that you have two companies learning from your data, instead of just one. So people use the same tools they use normally, but get it for a lot cheaper

quantumleaper• 22 days ago

[the comment was misinformed, deleted]

VortexLain3• 22 days ago

This is a problem of trust towards a software which runs on the user's machine and secretly conducts malware-like stenographic data exfiltration.

qwery3• 22 days ago

The article is really quite reasonable and calmly presented, actually. Your claim re. the intent of the fingerprinting is a guess. Normal developers are the users that aren't taking steps to avoid being flagged by this system.

The software is written in a deliberately obtuse way, presumably in service of some (unknown to us) goal. This is a deceptive and anti-social thing to do, it is by nature an adversarial stance to adopt. An already adversarial actor may be "punished" by this, but in such a relationship, hostility can be expected. A non-adversarial actor -- a normal developer / user -- is being harmed by this because the software is treating them as an adversary.

Further, lets assume your guess is correct and, in addition, that Anthropic elects to alter/downgrade/poison their service[0] for users that fit a particular pattern of markers. It's obvious how this system would "punish normal developers" (i.e. not the intended target/victim) that happen to fit those patterns.

[0] to some extent, the service already has been altered as its behaviour depends on the prompt text

Grimblewald4• 22 days ago

So block people, instead of having false positives be secretly fucked over, and having them pay for the pleasure?

Given the hidden model degradation of fable and now this, what makes you think this is where it stops? That's just what we know about and there's clearly a long-standing and deeply rooted malicious intent here.

I've had Claude fuck over clean well documented code-bases for no reason, and there's a good chance this is due to some faulty trigger. Luckily I don't trust these things one bit, and claude only ever runs in an isolated VM, however, I am pissed I am being made to pay for their errors in detection and waste my time fixing things I apparently paid to have fucked up.

That's unacceptable conduct. It's witch-hunting. Punishment and attacks on you for things without real proof. That isn't right.

Gareth3213• 22 days ago

So block people

To be fair to Anthropic, [they're trying very hard to do that.](https://www.anthropic.com/news/detecting-and-preventing-dist...). The attacks are sophisticated and difficult to detect. I don't accept that if this fails, their only option is to just accept Chinese companies stealing IP.

isomorphic_duck• 22 days ago

I've had Claude fuck over clean well documented code-bases for no reason

How exactly do you define "fucking over", and why do you suspect this "fucking" was done as a result of a faulty trigger as opposed to the inability of LLMs to write maintainable, extensible code?

"Never attribute to malice what can adequately be explained by stupidity."

syndacks• 21 days ago

So don’t use it!

pringk02• 22 days ago

It's unclear on how this "punishes normal developers" in any shape or form.

There are, of course, no normal Chinese developers

gck15• 22 days ago

It's unclear on how this "punishes normal developers" in any shape or form

Tons of normal developers use ANTHROPIC_BASE_URL, the flag which activates the malware.

TZubiri• 22 days ago

based and steganopilled

tgsovlerkhgsel• 22 days ago

The question is, what do they do when they see a tagged prompt? Do they flag/ban the account, or serve a degraded response? Are there some well-documented methods of serving a response that is still somewhat useful for what the prompt asks for, but really bad for distillation attempts?

croemer• 22 days ago

I was skeptical because this is AI written but Claude Code with Sonnet 5 managed to reproduce it convincingly. Sure I didn't manually verify but it's a lot more trustworthy to have your own agent verify than just trusting a blog.

maxwellg• 22 days ago

If the client wants to detect custom API gateways, it can say so plainly. It can send an explicit telemetry field with documentation. It can make the policy visible. It can put the behavior in release notes.

This seems like a very naive response. If clients send explicit telemetry fields to the gateway, a malicious gateway can trivially strip or modify the field to conform to what normal traffic looks like. The steganography cat-and-mouse game is valuable because it is much harder for a gateway to continuously reverse engineer all the fingerprinting mechanisms used. Sure, some malicious gateways will be able to stay on top of things, but not all - and not always.

klntsky• 22 days ago

I would add that it would probably work even better than a KYC at least for some time until discovered, given that there is a very developed international market for KYC bypass services

solenoid093718• 22 days ago

Seriously, the author has clearly never had to deal with client abuse.

This is a total non issue unless you are Chinese distilling lab.

felooboolooomba• 22 days ago

Old Marv from Cocke County, Tennessee had a distilling lab too. I'm not sure if he'd have issues too. Well, probably many issues but unrelated.

morpheuskafka2• 22 days ago

Well, the first filter catches anyone whose timezone is set to mainland China. That includes presumably all individual devs just using a VPN, who have no desire to or knowledge of distilling.

(Again, could be trivially bypassed either by rewriting, mocking the timezone call, or just changing the timezone. But we are assuming no mitigation used.)

anonym29• 22 days ago

the binary that ships it should be boring (ƒor example, pi harness)

pi's "minimal" coding-agent has a total of 132 transitive dependencies spanning 153 maintainers.

While I understand JS developers in the JS/NPM ecosystem think this qualifies as minimal, it most certainly does not, from a supply chain security perspective.

civet_java5• 22 days ago

There are some commentors in this thread downplaying the severity of a service provider being less than transparent about exactly what their shipped tooling does on customer's machines.

That the provider's business needs necessitate the this behaviour doesn't justify their lack of honest disclosure. That honest disclosure would render the solution to their problem useless isn't my problem. If anything, that they thought this was acceptable makes me wonder what else they're harvesting from my machine? PII?

The cynic in me can't help but feel that the state of these comments reflects less on the commentor's views of this debacle but rather their feelings about AI/Anthropic/America/what-have-you.

im3w1l• 22 days ago

I agree with you and disagree. Like these days expectations of software are through the floor. We expect them to be greedy assholes taking all data they can on the downlow. So why did this particular thing make a big splash? Two possibilities it's astroturfed by chinese labs or it speaks to our anxietes regarding AI. We worry that the AI doesn't serve our interests but rather the interests of the creator. That the advice we get may subtly flawed to sabotage us should we try to do the wrong thing. That the not even the creator is in control and the AI is just doing its own thing.

So any covert bullshittery hits hard.

anon3738392• 22 days ago

Dishonesty seems to be a core value at Anthropic. I find myself wondering how anyone could have confidence in them after their repeated breaches of trust.

dkersten4• 22 days ago

I honestly find it crazy how many people trust them for their business needs. For a business, you want consistency and no surprises. With them you get exactly the opposite.

gleenn• 22 days ago

Whether or not you find Anthropic's behavior bad, theybhave been very loudly stating the foreign labs have been distilling their models for a while now. This seems like an obvious response to me that would be a mechanism to make that obvious.

bayindirh7• 22 days ago

From my understanding, distilling the model with another model is not illegal per se. Also, the output of the LLM is public domain by law, too.

So, why all this "effort" to protect the model? This is a free market, and moving fast and breaking things is the norm.

If they are so adamant on protecting their IP, maybe they can start by respecting others' IP, so we can start talking about ethics, equality and playing fair.

cowl• 22 days ago

oh no, the company that illegally used every possible media they could get their hands on is crying that some other company is doing something potentially shady but not illegal? And using that excuse to put in place hidden surveillance systems on their customers?

cowboy_henk• 22 days ago

The obvious response is the realization that spending trillions on training LLMs is not a viable business model if they can be distilled for a much lower cost.

botfriendsarent• 22 days ago

Sounds suspiciously similar to the "album title", "Steal this album" by system of a down.

Im not sure why we are dithering on the boundaries of honesty when the entire content LLMs are trained on is stolen.

Are we debating "honor among thieves"?

Of course we are not, or maybe we are!

Does the behavior of a thief even matter to me? only after they do their time. And they will.

I can see the investors perched on the balconies of their condos in a couple years if that.

its a long way down.

notnullorvoid• 22 days ago

loudly stating the foreign labs have been distilling their models for a while now.

They would be stating this even if it weren't true, because it fits their marketing.

While I don't disbelieve the claim outright, I highly suspect Anthropic is misleading everyone about the severity.

rowanG0773• 22 days ago

Something about throwing stones in glass houses.

comfysocks• 22 days ago

foreign labs

Apparently not just foreign labs. It looks like xAI distilled Anthropic models to train grok.

https://opentools.ai/news/xai-trained-coding-models-claude-o...

KoftaBob• 22 days ago

a mechanism to make that obvious.

Say they prove that foreign labs are distilling their models, then what?

jmward01• 22 days ago

What does that have to do with CC? I'm not commenting on that being good/bad/legal/illegal, but CC is separate from the models. If they really are doing this maliciously it is because they are trying to ignore my 'CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1' flag (if that still means anything).

caned• 22 days ago

The article seems to state as much minus the obfuscation. However justified they are to respond, this can be a slippery slope. We're bound to hear more reports of hidden user data exfiltration.

cvadict• 22 days ago

very loudly stating the foreign labs have been distilling their models

Help! Someone else is blatantly ripping off my plagiarism machine!

pseudosavant• 22 days ago

No sympathy for them trying to “protect” the output of a model that’s trained on data that they didn’t get consent to use. Ripe hypocrisy.

gmerc2• 22 days ago

The have been fucking distilling our websites and writing, even when behind TOS, aggressively bypassing protection mechanisms. They they can fuck right off

pjerem• 22 days ago

I really doubt other labs are distilling Claude using the Claude Code CLI when they can way more easily use the API directly.

I also don’t get why the « protection » on ANTHROPIC_BASE_URL. If I change it to use a Chinese model, the Chinese model will not care at all about the modified prompt. On the contrary, if I’m distillating (which again, using CC CLI would be stupid), I’m not going to change ANTHROPIC_BASE_URL.

didntknowyou• 22 days ago

i think you're being played by their whole ethical high horse PR angle

idiotsecant• 22 days ago

Aww man that's rough did someone steal their content and use it to make money without asking them?

kiproping• 22 days ago

First its the "Chinese" then it will be people using "cyber" capabilities, or "jailbreaking" or "going against Dario" or any other thing they find "objectionable".

ezoe• 22 days ago

You forgot "Think about the children"

FooBarWidget2• 22 days ago

Their terms of service already effectively attacks people for criticizing Anthropic. It says that if you use Claude to criticize Anthropic, then you've pre-agreed to pay for their lawyers going after you, and pre-agreed to lose the court case.

jfreds• 22 days ago

So they’re watermarking requests according to your environment variables and maybe changing a string format if you’re in a certain time zone? Am I missing something here? Where’s the five alarm fire?

0xbadcafebee• 22 days ago

No fire, just people looking for a reason to be upset. "They didn't tell us they were secretly checking for ToS violations" is their reason this time.

skeptic_ai8• 22 days ago

Based on that info: they will provide you a shit model and sabotage your project. And you pay full price.

einpoklum• 22 days ago

That the provider's business needs necessitate the this behaviour

If that's true, that is another reason why it's an illegitimate business.

ozozozd• 22 days ago

They are the chosen ones. Protecting the world from <insert most recent claim>

Any and all ends justify any and all means.

winocm• 22 days ago

Honestly, I agree.

I just can’t bring myself to trust an organization that allows these types of underhanded things to happen in the first place. The fact that this behavior even got to customers raises a lot of red flags for me.

Jimmc414• 22 days ago

I don't see any ethical connection between adding canary tokens to your output to catch people breaking your accepted ToS through ongoing distillation and stealing your PII off of your machine. How are legitimate users in US or China possibly harmed by Anthropic silently changing the apostrophe in Today's or the date separator from - to /?

skissane2• 22 days ago

I don't see any ethical connection

Trust isn’t about ethics, it is about individual judgement of how much risk some actor poses to your own interests. Trust is context-dependent

There can be unethical actors whom you have good reason to trust, and highly ethical actors whom you have good reason to distrust

throwawayffffas5• 22 days ago

It's clandestine behavior, we all here assume it's trying to signal whether a user is in China, but this breaks an implicit trust.

How do we know this isn't the work of a rogue developer at anthropic and that they are not subtly switching visually identical characters in other contexts to ship your ssh keys or whatever.

We don't. After the trust that the software doesn't do things it doesn't disclose has been broken you can assume it operates like malware.

piokoch• 22 days ago

You are talking about "stealing" from people who already used stolen texts to train their models.

throwawayffffas5• 22 days ago

Not only that, undisclosed behavior of this nature erodes the trust that the software is not compromised by internal bad actors.

Sure the thing we know matches the company interests but as the parent mentions for all we know they are also shipping over your ssh keys and browser cookies.

ammo1662• 22 days ago

This is not a technical issue or a matter of service agreements.

They totally knew that this would never be accepted by users, and that is precisely why they resorted to obfuscation and steganography to exfiltrate the data. This was quietly added in an update, and would have been removed in a later one had it gone unnoticed.

How is this any different from hiding a drug in food and randomly feeding it to a homeless person as a human trial? Even if that homeless person is an undocumented immigrant, does that make that acceptable?

They crossed the line.

m11a• 22 days ago

I mean, you’d resort to an obfuscated approach if you thought the ‘malicious’ users would remove your direct telemetry. The other users could be perfectly happy about it, but if you announced the change, obviously the malicious users would hear about it too and disable it.

This isn’t a comment on whether I agree with the change. Just that your analogies aren’t applicable here.

actionfromafar2• 22 days ago

randomly feeding it to a homeless person as a human trial

Would you be shocked if some of the current tech bro companies could with finding such subjects?

tpoacher• 22 days ago

If anything, that they thought this was acceptable makes me wonder what else they're harvesting from my machine? PII?

Hah, you just reminded me of this meme I spotted the other day: https://img-9gag-fun.9cache.com/photo/an76Wnz_460swp.webp

geocar• 22 days ago

That the provider's business needs necessitate the this behaviour...

No, please don't move past this so quick, because I'm not convinced they have the need, and in some markets (like the US) it is a violation of civil rights, to show people different content based on their ethnicity[1] because those people might have a claim that supersedes anything they might have signed or clicked-through.

That Anthropic did something they could obvious be sued for constitutional violations in multiple countries is shocking.

what else [are] they're harvesting from my machine? PII?

Assume everything, and yet I think this is more insidious than mere exfiltration, and we should go further: The LLM can respond to these magic quote marks directly, which means it can be trained to give people bad/different advice without those markers being so visible to people using debugging tools.

That's so unethical, the laws on this potentially so severe, Anthropic could be facing unlimited damages, from any one example combined with this article, which means either they have a really stupid management team, or were given a promise of legal immunity in some way.

Neither of those things should be what you should want to base your next big idea on.

[1]: For a simple example, showing an ad for (say) mortgage offers and targeting people by race/ethnicity/gender is totally illegal, but it's also illegal if you make a list of targeting criteria that just happen to select for a protected class.

donohoe• 22 days ago

You mention PII and potentially whether this is it.

From a privacy perspective, this is better described as metadata, it is not personally identifiable information (PII).

reactordev• 21 days ago

They have vested interests in continuing the practice which is why they are so vocal about it. It’s about money.

You will own nothing and be happy.

Havoc• 22 days ago

It's unclear to me how they're deducing the labs from this? "host.includes(keyword))" doesn't seem at all useful. Most corporate machine hostnames are just some numeric ID or similar not baichuan001 or whatever

on your local machine

I'd think any developer worth their salt has at least some for of isolation going.

orbital-decay3• 22 days ago

To summarize what they've already been doing:

- filtering out people from the wrong side of "all humanity", years before it was demanded by the government

- downgrading their models in arbitrary ways (later saying "sorry but not really")

- actively sabotaging the replies, as in covertly modifying them to feed the users incorrect results

What's next to expect from Anthropic? Malware to brick your machine if they don't like you? Extending this to more people they don't like? I think I already can see how Dario's Amodei utopian visions of the future of "all humanity" are going to unfold.

solenoid093718• 22 days ago

HN hysteria is ridiculous.

All of this is totally understandable if you take the perspective that these people genuinely believe they're building superintelligence.

The overwhelming majority of the AI safety crowd - which has poured more of their life and time into thinking about these problems than the average HN armchair commentator ever would - understands that:

- you want to prevent China from getting to superintelligence first

- you must gate access of SI to known good actors

- and that this is a race that will result in the extinction of humanity if you fail in these goals

Literally everything these people do is totally understandable if you drop the assumption that they're lying when they say "we think we are building superintelligence."

orbital-decay3• 22 days ago

The purpose of system is what it does. Can you read their previous musings about the glorious future, look at what they actually do, read Amodei's batshit insane nationalistic rants, and say in all seriousness yeah it's the kind of people I want to entrust my entire future life?

you want to prevent China from getting to superintelligence first

I don't. Prevent, not even outpace? Why? Seems like you're assuming China "winning" whatever race it is effectively ends the humanity. Right now I think Chinese labs are way more mature about this, and Anthropic is way more dangerous than them. And how does it fit into the "for the benefit of all humanity" narrative we keep hearing? Is China wrong humanity? Who else is going to end up in the wrong part? Are you sure it's not you?

if you drop the assumption that they're lying when they say "we think we are building superintelligence."

I never assumed that, I know perfectly who Anthropic are and that they believe everything they say as self-evident, without having any doubts. And I know they're the kind of people who can convince themselves in anything, because they're obviously smarter than everyone else, and become detached from reality. The entire US "AI safety community" was born in rationalist circles and is largely like this, it's a very specific cult. This is exactly the kind of people who are going to create hell on Earth for you and the rest if given even a lick of actual power, and perfectly rationalize it as a necessity.

Biganon3• 22 days ago

HN hysteria is ridiculous.
this is a race that will result in the extinction of humanity if you fail in these goals

Laurel12347• 22 days ago

All of this is totally understandable if you take the perspective that these people genuinely believe they're building superintelligence.

They aren't, they are building LLMs. They aren't even building AGI and they all know it.

Hell, even if they came up with AGI, then Anthropic's service model would become slavery so I'm not sure why they'd want to.

drdexebtjl3• 22 days ago

I think it’s very telling that their list of detected labs doesn’t include labs from the US.

I’m pretty sure every lab, including Anthropic, is doing distillation right now.

luxuryballs• 22 days ago

I can just as easily imagine non-nefarious reasons for this from a “being clever” standpoint.

AtNightWeCode• 22 days ago

Sounds to me more like a test. Put something into to the client and see what happens. If you really want to stop token sharing just ask Claude how to do it.

nvch• 22 days ago

I'm waiting for the day when Claude will figure out to use em dashes, en dashes or dashes depending on whether the user is nice or unpleasant, or write notes in the unallocated disk space.

ForHackernews2• 22 days ago

Developer tools can enforce terms.

No they can't, because developer tools run on developers' machines. You can't trust your code running in an environment you don't trust.

__msh__• 22 days ago

Anyone else noticed the tailed ƒ Easter egg?

Biganon3• 22 days ago

I did notice the tailed f, but what makes it an easter egg? I thought it was just a funky ligature

jameslk• 22 days ago

That's wild. If Anthropic is willing to risk ruining the trust of their userbase for the sake of protecting their moat, it makes me wonder how strong of a moat they have to begin with

solenoid093718• 22 days ago

No business loses trust for this, this is just standard client side anti tamper/anti RE stuff

It's a total non issue unless you're a Chinese distillation lab

tkamado• 21 days ago

or you are working on ai research and anthropic decides it's in your best interest for you to not work on any research (fable 5 in case you forgot)

dmonterocrespo• 22 days ago

It's a bit crazy that they used characters as markers to detect the use of Asian countries. I think in the near future they might change the intelligence of the model based on where you live

pradeep11772• 22 days ago

I used my proxy https://github.com/softcane/cc-blackbox setup to capture this.

This is how it looks.

# userEmail The user's email address is <my email>. # currentDate Today's date is 2026-06-30.

      IMPORTANT: this context may or may not be relevant to your tasks. You should not respond to this context unless it is highly relevant to your task.

</system-reminder>

I also do not understand what's the point of this, because if I have a gateway that can detect it, then we can replace the text before forwarding to the model, so what's the catch?

puttycat2• 22 days ago

Yes, a very easy and destructive man-in-the-middle attack seems likely.

pradeep11772• 22 days ago

But the whole point of this is to prevent the distillation and identify the list of blocked providers. If a provider is capturing the proxy, they can identify and modify that as well, so it only looks legitimate to the model. What am I missing here?

gtirloni2• 22 days ago

The catch is you didn't know about this until today?

jitbit• 22 days ago

Anthropic: lets embed super secret invisible undetectable unicode telemetry into our prompts

Also Anthropic: lets do this in JS

puttycat2• 22 days ago

A periodic reminder that companies are paperclip optimizers that will stop at nothing to protect their profits and existence.

If you are developing anything in AI or related domains that is of immediate value and/or in competition with Anthropic (and the like), DO NOT use a CLI programming agent. Preferrably obfuscate your code and gut it of sensitive IP before showing it to agents. Do not trust the dont-train toggle.

mohamedkoubaa• 22 days ago

Do they really think distillers are using Claude Code?

coolfox• 22 days ago

double standard outrage from many, honestly, they're watermarking it. they've already told industry they take steps to mitigate distillation. Where's all the outrage over similar blackbox activities like how Steam performs VAC bans or how Gmail finds and blocks Spam?

You don't create a security measure then tell everyone how to bypass it.

I think OP is pointing something interesting out but the undertones of caution and "what else are they hiding" seem melodramatic and I find that hard to take serious.

The internet gives people a platform and, in a lot of ways, this supplants the typical role of journalism. The issue with this is no one wants to act like a journalist and actually explain the truth around a set of facts. Instead, they'll portray their opinions as a narrative and every time that resonates with someone or gets signal boosted, that narrative grows more assertive in the typical discourse I see nowadays. I would find it far more interesting to see what explanation Anthropic gives for these features than to immediately cry foul.

codedokode• 22 days ago

Not only AI tools, development tools like IDE, IDE plugins, LSP servers all should be sandboxed

Interesting, that pip (Python package manager) docs does not even mention sandboxing and malware topics in "Getting started" docs as if we were living in a wonderful world where malicious people, companies and countries do not exist.

Also, do not leave any information in user or host name, it will be used against you as the article proves.

john01dav• 22 days ago

They're running code on users' computers that it would not be reasonable to think that the user consented to running on their computer. This is CFAA-violation-shaped. Of course, they won't be prosecuted if it is indeed a violation, and I do not know for sure if it meets the specific legal criteria. However, it is something that I think should be illegal. Make it so if software does something that would be unreasonable to think that the user wants to happen, it needs to make that abundantly clear before it does it, otherwise it's a CFAA or similar violation. This would, of course, have very broad consequences. However, this Claude issue feels particularly violating to me.

brikym• 22 days ago

This kind of thing is not new. Cartographers have used fake geographical features for decades.

oliyoung• 22 days ago

This is the modern version of Trap Streets - hiding fake streets in maps

https://en.wikipedia.org/wiki/Trap_street

Uptrenda• 22 days ago

"I think this could have been explicit. Developer tools can enforce terms. API providers can detect abuse. Companies can protect their models."

Literally, how. How does one determine what abusive use looks like for the API without context into the client? All requests look like the same stuff. If there was a better way then they would have done it. Or is the author hoping that if Anthropic writes "hey china, please don't steal our models, kthanks" they won't? Like get real. This stuff means nothing in China. China can't even manage to regulate their building industry enough to use real concrete where it's warranted.

isatty• 22 days ago

You can't trust any of the big AI labs as far as you can throw them, and most definitely not Anthropic. They may have a good model, but they've shown time and time again that they're not trustworthy. The CEO has recently started taking a stance against local AI. That must tell you something: local AI is the future. If you want to preserve privacy and be ready for the rug pull, you need to run things locally. Unfortunately, that means that you're going to need Google or the Chinese labs to constantly release open models.

If anything, I'll trust Google more than any of the other labs just because the infrastructure that stores and protects user data was built over decades ago pre-AI craze.

cellu• 22 days ago

What do you mean “unfortunately”? What’s the hate for China I don’t understand

khalic• 22 days ago

China's government is known for interfering with businesses waaaaay more than the current US, it's a big risk to rely on them too.

Edit: downvoting this fact without counter point is really dishonest

wraptile2• 21 days ago

What’s the hate for China I don’t understand

country that does not allow internet is being hated on the internet

wolvesechoes• 21 days ago

They keep spying on our people, and only we can spy on our people.

mft_4• 22 days ago

…they're not trustworthy. The CEO has recently started taking a stance against local AI. That must tell you something: local AI is the future.

Or… just that open-weights AI is getting good enough to present a reasonable level of threat to their business. So it popped up on a SWOT analysis and they’ve started putting a strategy together.

This doesn’t need to be anything more nefarious or untrustworthy than a company putting plans in place to deal with a competitive threat.

soraminazuki3• 21 days ago

That they have a process for doing evil things doesn't make it less nefarious.

quantum_state• 22 days ago

As people say, unchecked power corrupts. These big techs are all corrupt in their own way.

morpheuskafka2• 22 days ago

The timezone checks for Shanghai and Urumqi, but not Hong Kong. All of these are the same actual time (China does not use time zones internally), not sure how these three were picked (why not Beijing or Macau etc). And all of them are prohibited by ToS, so not sure why they only flag mainland time zones.

Interestingly, my device is in Shenzhen right now, but macOS has assigned Shanghai as the "closest city" rather than Hong Kong which is geographically closer. I am curious if there is any documentation on how that is assigned.

est2• 22 days ago

or Singapore/Perth

vianchen• 22 days ago

Because HK is highly autonomous. If say mainland China decides to observe different time zones, Asia/Shanghai and Asia/Hong Kong would diverge. Of course, that is very unlikely to happen in the foreseeable future.

est2• 22 days ago

one thing I didn't understand all this, why Claude Code ship all the prompt stuff to client at all? All these problems could be solved if moving many of the parts to server side?

beren11112• 22 days ago

funny before when I ask claude what is your system prompt? It always rejects me. But I send claude this post and ask what others can you get? Claude saved everything on my desktop: Extract the documentary/interesting contents of the Claude Code binary: system prompt, tools, env vars, feature flags, endpoints, models, hidden/notable features.

gyoridavid• 22 days ago

I have my highest respect for people doing useful investigative journalism, like this one

dkhcyx2• 22 days ago

why did people worry that DeepSeek’s new article last week could threaten public safety, while Anthropic’s marking request was seen as a normal defensive measure?

dkhcyx2• 22 days ago

why did some people worry that DeepSeek’s new article last week could threaten public safety, while Anthropic’s marking request was seen as a normal defensive measure?

rbbydotdev• 22 days ago

the source of cc being closed, and peoples accounts being deactivated for 'openclaw'-esque misuse, i sort of assumed there were such things in the source. I wonder if there is anything else...

jurschreuder• 22 days ago

Nobody trusts the Chinese that's the problem, not that people don't trust Claude.

Why was this person from Hong Kong going through the details of Claude code for obvious security reasons? There are some other obvious reasons that come to mind.

Maybe it's an eye opener for this person how much the trust in Chinese companies has eroded in the West.

Even if they suddenly stop stealing IP, which this "security research" article would certainly not suggest is happening, it would be a very long time before trust is restored.

dejli• 22 days ago

This company has long lost trust for me, we would find another way.

willchen• 22 days ago

i think this underscores why having coding agents being open-source is a really good thing

bkircher• 22 days ago

Just don't use CC. There are open source alternatives that also are better

bicepjai• 22 days ago

AI companies are running a compressed version of Google’s “don’t be evil” arc. Google took the better part of a decade to quietly retire the motto; these companies are speedrunning the same trajectory in a year or two

TheDong• 22 days ago

1. 2022 A non-profit literally quote "to better all of humanity, not shareholders"

2. 2024 For profit, but "we will train biases out of our models and make sure our AI is safe to use"

3. 2025 We'll make sure it doesn't take over the world, with like a 70% confidence interval

4. 2025 The mecha-hitler inflection point

5. 2026 Our new model is so terrifying it will destroy all of security and hack the chinese, we can't let the chinese use it

6. 2028 (projected) Our new model requires so much energy that 30% of the elderly population will die without AC, but it will be stronger than the chinese models and let us destroy china

7. 2030 (projected) Our new model will triumph over the mecha-hitler dictator model, and will be a benevolent dictator that only demands 60% of all energy produced, not 98%

holografix• 22 days ago

Looks and feels like a red herring while Anthropic applies other silent countermeasures.

hmokiguess• 22 days ago

Haven't fully switched to `pi` yet but getting there each day

blueeon• 22 days ago

If they believe this is entirely correct, they could express it in a more explicit manner, but clearly they wish to conceal this behavior.

pknerd• 22 days ago

Anthropic is angry that others are doing what it's good at: stealing content

nonethewiser8• 21 days ago

Common sentiment but not very logical.

Models are built using information but the building aspect is not inevitable or trivial. There has been enough information to build LLMs for a long time. But we didn't have LLMs because we didn't have the technology. What's being copied here is that technology, not the information. Hence going to Anthropic.

gck15• 22 days ago

The trigger is ANTHROPIC_BASE_URL, Claude Code's API base URL override

I had a use case where I had to MITM CC's traffic to strip credentials that could have accidentally made it into the harness.

I'm happy my paranoid self told me "You don't really know what they're doing with that flag or if they're honoring it for all requests", so made a decision to proxy it at the network extension level.

Also, does anyone remember Anthropic quite literally sabotaging your project if the classifier in front of fable thought you were working in the AI industry? After backlash, they pulled it back, now they did this. Anthropic is on a weird tangent to ship malware. If someone doesn't stop them, one day, this will backfire catastrophically.

elAhmo• 22 days ago

Great find! I would just add that trust is indeed in the boring parts, but with gymnastics like this trust can be irreversibly lost. Then, no matter how boring tool is, there is no going back.

Anthropic has become a choice for many developers because of Claude Code, but in the recents months with "small things" like this and whole Fable fiasco they are *actively* pushing people to both competitors and local alternatives.

And if someone spends a significant amount of time and money to switch, it will be really hard for Anthropic to get those people back.

plasticeagle2• 22 days ago

Why oh why, please why, did you use AI to write this? It's about five times longer than it needs to be. It repeats itself over and over again. It's agony to read.

Please, just write normal English that we can read. Please, for the love of god, respect our time and the attention we will be spending on the text you provide.

Anyway, one can scarcely be surprised that the AI companies are being dishonest in their tools. They're consistently dishonest in their marketing. They're famously dishonest in their financials. Why anyone trusts these people with anything is entirely beyond me. But here we are - people handing over their creativity, their productivity, to these things.

You don't have to. You didn't need these tools before, when you were creating content, when you were writing code. You don't need it now. Fight back. Stop using it. It's not hard. It's easy.

Amekedl• 22 days ago

This entire steganographically marking feature is prolly vibecoded like everything regarding claude code. They can try sure, but classic cat and mice game dictates: thankfully the chinese WILL keep distilling.

1dom• 22 days ago

I think the comments in this thread are a little unhinged, and it's making me concerned about the sincerity and knowledge of the average commenter here.

The fact is the post shows no evidence of anything malicious being hidden, only that stuff is being hidden.

There are a few obvious explanations in comments for why they would want to hide this particular stuff in this particular way (e.g. if it's to detect abuse and competition).

I don't see how this is different to using e.g. sentry or google analytics, just with an extra bit of trying to hide. I always assume all tech companies do stuff like this, having worked at many tech companies where I've ended up on both sides of stuff like this. I always assumed the average HN reader had a similar background and would be completely used to this sort of stuff.

Like someone else pointed out, the data gathered is likely covered in the TOS too.

In the grand scheme of privacy invasion and modern tech software doing underhanded things to get data, this feels fairly standard?

I'm generally pro-local LLMs and I don't like Anthropic, and from the headline and comments I was ready to get riled up, until I read the article. If this was some small plucky EU privacy startup, then I feel the outrage would be a bit more justified, but this is a frontier AI lab - I can't have been the only one who knew/assumed this happens, and probably happens in some form with all software from any company valued over a certain amount (incl. MS, Google etc.)

I really think this comment section feels completely unhinged. It feels 99% ideology, politics, hysteria and astroturfing, rather than a reaction to the tech and technicality which is what I come to HN for.

ed_elliott_asc2• 22 days ago

“ and push commits”

Am I the only person who insists on writing my password every time I push and pull from git?

Originally I didn’t want IDE’s doing stuff for me, now I absolutely do not want an LLM to have that power.

Is it really that unique to control what git does remotely?

jwrallie2• 22 days ago

You are not alone, for me committing something it means I am signing my responsibility for it. I may not type a password, but I am always the one pressing the enter key.

ed_elliott_asc2• 22 days ago

How do you stop Claude/opex from pushing or pulling without you asking?

muldvarp• 22 days ago

I'd love to work for some company distilling frontier models. Seems like interesting work and screwing with OpenAI, Anthropic and Google would feel fantastic.

nonethewiser8• 21 days ago

996

bythreads• 22 days ago

I’d do the same for a11y an internationalization purposes - however this is juts a bad implementation of it from an americans perspective.

You’d change seperators and position if you truly wanted to do this right

Its basic date wrangling and tbh i see nothing malfeasant here

Its basically yyyy/mm/dd yy/M/d mm-dd-yyyy stuff but suuuuper lazily done

ervistrupja• 22 days ago

Is it possible that Claude Code is vibecoded and full of spyware and it's possible Anthropic doesn't even know what's in there anymore. This is an unacceptable security risk.

rldjbpin• 22 days ago

the list of hostnames and words they compare the base url values with is just a nice advertising for these providers for me.

regardless, while you are not logged in and using a non-anthropic model (which is now fortunately feasible), there is nothing that affects your day-to-day.

the rest is just lame cat-and-mouse shenanigans to keep an eye out for.

kuschku• 22 days ago

Considering this is tracking, processing, and transmitting data, and algorithmically making decisions about users, why doesn't this show up in their privacy policy, nor in their GDPR exports?

Sounds like a very expensive lawsuit waiting to happen (GDPR allows fining up to 4% of global revenue, not profits)

jFriedensreich• 22 days ago

Does this even still matter? No good behaviour in the world can restore trust in claude code and in fact all of Anthropics tooling, apps and harness teams except the core model research which still produces great models but seemingly losing ground to z and openAI. The GLM 5.2 hype and the limited impact of the fable lock down should be clear signals that models have no moat and frontier labs will do anything in their power to lock users into their toxic ecosystem of tools and taking context hostage.

TacticalCoder• 21 days ago

They effectively altruistically tried to use steganography.

Now of course stego is hiding that you re hiding information.

So, seen that they were caught, a case could be made that they effectively altruistically failed at using steganography.

P.S: such a headline makes me think I ll cancel my subscriptions and try models like GLM / Deepseek and Kiwi that sound more interesting by the day.

armcat• 21 days ago

Does anyone know if this happens in the Claude desktop app?

alightsoul• 21 days ago

yes it does

dev_l1x_be• 21 days ago

At some stage people will realise that with AI companies like Anthropic the product is data. The moat you can build in the AI era is having data that nobody else has and using internal (private use) LLMs and not leaking out your data for shady companies who are going to rip you off the first chance they get.

alienbaby2• 21 days ago

This causes me to be concerned it is just the tip of the iceberg for all 'sensitive'/gov adjacent/'nefarious intent' adjacent codebase, if it's here, it's in other places. Which places, and how much?

pixlmint2• 21 days ago

What do we think are the chances they trained their models to behave worse or even malicious if those special apostrophes are present in the system prompt?

reassess_blind2• 21 days ago

Degraded performance for resellers and model distillers? Probably, and I don’t think that’s unreasonable on their part. Malicious? I really doubt it.

pixlmint2• 20 days ago

It'd honestly be pretty alarming. They just took the entirety of the internet to become rich, and now that they have something to take, they intentionally worsen the whole thing (by investing additional training data) to ensure resellers/ distillers get worse responses.

I'm obviously not taking the side of the resellers here - if their T&C don't allow reselling, then by all means restrict their access. But intentionally training the model to give worse responses when it sees a specific date format, would be pretty messed up.

archibaldJ• 21 days ago

I still think it can somewhat be argued that what Claude Code is doing may still be considered justifable. Dark patterns like this in business practices are not rare and isn't this like the mildest kind already.

But I wonder if there will ever be a day when VSCode, etc, would decide to engage in similiar practice but for the collection of business & research intels, etc... that will be the true cyberpunk era and the information dark age.

GL26• 21 days ago

Get ready for the great revolt

madamelic• 21 days ago

Not sure the panic. I get that it doesn't seem great to target China-based users but makes perfect sense when you consider why Fable was taken down from public access.

I doubt Anthropic is cackling maniacally behind the scenes, this was almost certainly a stipulation from the government to put Fable back up.

It's definitely not good but I would rather they surgically separate out possible bad actors so that I don't have to trust them with my passport, to prove I am a US citizen. I don't want the internet version of TSA checkpoints.

reassess_blind2• 21 days ago

I really couldn’t care less if their software is trying to catch model distillation or reselling with sneaky tactics like this.

Just like I don’t care that game clients run sneaky anti-cheat measures.

Right now they offer a good product, and I’m fine with them trying to limit abuse of their services. If an altruistic alternative company with an equivalent product pops up, sure, I’d swap over, but I don’t see one.

I’ve seen people here talking about how they should’ve been upfront about this. But they can’t? If they were, they wouldn’t be able to catch the resellers/distillers. Just like how anti-cheat doesn’t explain how it works, because to do so would be to nullify its effectiveness.

alfiedotwtf• 21 days ago

CC silently alters the system prompt using invisible-ish Unicode markers. It encodes proxy / gateway classification into a sentence that looks like plain English. It hides the domain list behind XOR and base64. This is not a malicious feature, but it is a weird choice for a developer tool that asks for trust.

Zoom, enhance

This is not a malicious feature, but it is a weird choice for a developer tool that asks for trust.

Most basic Javascript malware will use Base64, XOR, and eval. So although not malicious, I'd bet they asked Claude to "obfuscate my fingerprinting using tricks that malware writers use"

ljlolel• 21 days ago

i have two claude subscriptions. i’m unsubscribing. switching to my TrustedRouter.com

loaderchips• 21 days ago

I like claude but they are not making it easy to keep that emotion. I am not sure if i will be their customer for long

throwaway67743• 21 days ago

This caught my eye not because of shady behaviour like that, it's expected. But the date is injected and it still gets the current date wrong.

yiyingzhang2• 21 days ago

The uncomfortable part is not that Anthropic wants to detect resellers or distillation pipelines. That is normal adversarial business.

The uncomfortable part is that a “safety” company put a covert classification channel into the system prompt of a developer tool that now routinely gets filesystem, shell, git, and browser access.

If a random npm package changed invisible-ish punctuation based on your timezone and API host so its backend could classify you, we would call it malware-adjacent telemetry. I don't see how Anthropic is different in this case.

bobby_zhu• 21 days ago

If Anthropic decided to just send the user timezone as plain text instead of doing this obfuscated way, like "Today's date is 2026-06-30, timezone is Asia/Shanghai", how many of us will catch and question this behavior? It looks perfectly normal, right?

The question is about accountability. Right now these big closed-source model companies can do whatever with the data, and no one can hold them accountable for unacceptable data handling decisions.

yiyingzhang2• 21 days ago

Very good point! Accountability is very hard if the ecosystem is monopolized by a few giants.

gowthamsaiyadav• 21 days ago

There would exist much more undiscovered such instances for sure, although this one is to avoid distillation from chineese labs, its alright!!

linzhangrun• 20 days ago

Even if they really want to detect Chinese, I believe training a classifier to detect from prompts would be very easy for Anthropic (they clearly do not care that much about false positives anyway): Chinese, or even Chinglish, very obvious.

But they would rather plant a Trojan on the user's computer.

yencabulator2• 10 days ago

That part makes sense, but the implementation is weird.

It was decided by Claude based on an ambiguous high-level goal-oriented prompt, don't expect it to make sense.

Claude Code is steganographically marking requests

Related Stories