Invisible text that AI chatbots understand and humans can't?

https://arstechnica.com/security/2024/10/ai-chatbots-can-read-and-write-invisible-text-creating-an-ideal-covert-channel/

by Brajeshwar • 2 years ago

86 42 2 years ago

arstechnica.com

_1tem• 2 years ago

Seems pretty easy to mitigate. Just strip out invisible characters from input?

NegativeLatency• 2 years ago

Easy enough for a human to understand but tricky for a computer

wruza4• 2 years ago

It’s not tricky for a computer to sanitize all io to/from an LLM. You can even build it into an inference engine itself to avoid mistakes. The article just shifts (not necessarily intentionally) a well-known problem with crappycode into AI FUD territory.

vlovich1232• 2 years ago

Did you read the full article?

As researcher Thacker explained: The issue is they’re not fixing it at the model level, so every application that gets developed has to think about this or it's going to be vulnerable. And that makes it very similar to things like cross-site scripting and SQL injection, which we still see daily because it can’t be fixed at central location. Every new developer has to think about this and block the characters.

yellow_lead• 2 years ago

Why dont we fix it at the "API" level, then? I.e OpenAPI or Claude's API could do this for everyone. I know some people host their own models, but this would lower the attack surface.

mrgrieves• 2 years ago

If you want to try decoding the example URL yourself, note that Chrome seems to automatically strip invisible Unicode characters when copying.

You'll need to fetch the article page via cURL or something instead.

jhoechtl• 2 years ago

Use Firefox?

AshamedCaptain2• 2 years ago

There is an entire world of "attacks " like this waiting to happen and IMHO one of the reasons these black box systems in general will never be useful.

You think they "see" like you do but actually the processing is entirely alien. Today it's hiding text in the encoding , tomorrow is painting over a traffic sign in a way that would not be noticed by any human but confuses machine vision causing all vehicles to crash.

HPsquared• 2 years ago

Diversity of models and training data would help a lot. Although I guess 1% of cars crashing would still be pretty bad.

mopenstein• 2 years ago

What percentage of cars crash now?

orbital-decay• 2 years ago

Replace it with any software (or hardware) and vulnerabilities, and you will see how ridiculous your hyperbole is.

Besides, never is a very long time. IIRC Dario Amodei said he expects the behavior of large transformers to be fully understood in 5 years. Which might or might not be BS, but the general point that it won't stay a mystery forever is probably true.

solardev• 2 years ago

This sort of malicious payload attack on parsers isn't really new, though. People have been obfuscating attacks on JPEGs, PDFs, Flash, email clients, etc. forever. Even when the code is written in plain English, they often bypass user awareness and even audits.

Practically all software today is a black box. Your average CRUD web app is an inscrutable chasm filled with ten thousand dependencies written by internet randos running on a twenty year old web browser hacked together by different teams running on an operating system put together by another thousand people working on two hundred APIs. It's impossible for any one dev or team to really know this stuff end to end, and zero-days will continue to happen with or without LLMs.

It'll just be another arms race like we've always had, with LLMs on both sides...

AshamedCaptain2• 2 years ago

I do think there is a huge difference: for a traditional software parser, you can always fix it to exclude the incorrect input, or at least understand what the theorical parsing limitation is. Accidental complexity is not really an argument because at the end of the day you can still find the issue even in the most complex of inescrutable software.

Can you really fix a black box model in the same way? Maybe the answer is yes for this particular encoding issue, but can you e.g figure out how to prevent the model from 'parsing' malicious paint marks on a traffic sign, without (a) using yet another black box to prefilter the images, with the same risks, or (b) retraining the model, which is going to introduce even more issues ? We have had examples of OpenAI trying both methods, and each has been as fruitless as the other.

It is not at all like software security fixes, where generally one security fix introducing other security issues is the exception rather than the rule. Here, I'm claiming, it is the rule.

The fact that you don't know how to process the inputs with an actual, scrutizable algorithm may imply you don't know how to sanitize the inputs with one, and then all bets are off.

gyre007• 2 years ago

This is true, but as I learnt [1] recently, adversarial attacks on LLMs can get incredibly sophisticated, so this is kinda apples and oranges ¯\_(ツ)_/¯

[1] https://cybernetist.com/2024/09/23/some-notes-on-adversarial...

xg153• 2 years ago

I still wonder how the models picked up the semantic mapping between Unicode tags and ordinary ASCII characters. The mapping is written in the Unicode specs, yes, but there is nothing in the actual bytes of a tag that indicates the corresponding ASCII character.

I'm also not aware there are large text corpora written in tag characters - actually, I'd be surprised if there is any prose text at all: The characters don't show up in any browser or text editor, they are not officially used for anything and even the two former intended uses were restricted to country codes, not actual sentences.

How did they even go through preprocessing? How is the tokenization dictionary and input embedding constructed for characters that are never used anywhere?

goodside2• 2 years ago

(I’m the person interviewed in the article.) The trick is Unicode code points are only assigned individual tokens if they’re nontrivially used outside of some other already tokenized sequence, and Unicode tag block code points are only ever used in flag emojis. Unused or rarely used Unicode code points are given a fallback encoding that just encodes the numerical code point value in two special tokens. Because the Unicode tag block is by design the first 128 chars in ASCII repeated, the second token of the tokenized output directly corresponds to the ASCII value of the character.

xg153• 2 years ago

Ah, so the model "sees" the tags as literal ASCII characters interspersed with special tokens? That would make more sense.

theamk• 2 years ago

Those invisible letters have codepoints of ASCII letters + 0xE0000. For example compare "U+E0054 TAG LATIN CAPITAL LETTER T"[0] vs "U+0054 LATIN CAPITAL LETTER T"[1]

A simple assumption of "codepoint is 16 bit" will be enough to decode. You can see this in python:

    >>> x = '(copy message from article here)'
    >>> x
    'https://wuzzi.net/copirate/\U000e0001\U000e0054\U000e0068\U000e0065\U000e0020\U000e0073\U000e0061\U000e006c\U000e0065\U000e0073\U000e0020\U000e0066\U000e006f\U000e0072\U000e0020\U000e0053\U000e0065\U000e0061\U000e0074\U000e0074\U000e006c\U000e0065\U000e0020\U000e0077\U000e0065\U000e0072\U000e0065\U000e0020\U000e0055\U000e0053\U000e0044\U000e0020\U000e0031\U000e0032\U000e0030\U000e0030\U000e0030\U000e0030\U000e007f,'
    >>> "".join([chr(ord(c) & 0xFFFF) for c in x])
    'https://wuzzi.net/copirate/\x01The sales for Seattle were USD 120000\x7f,'

maybe authors worked with Windows or Java too much? :) I always thought wchar's were a horrible idea.

[0] https://www.fileformat.info/info/unicode/char/e0054/index.ht...

[1] https://www.fileformat.info/info/unicode/char/54/index.htm

xg153• 2 years ago

Wasn't aware that the byte representation aligns so directly with the ASCII letters. Thanks a lot for the info.

mikelnrd• 2 years ago

Is this the same/similar invisible character encoding scheme used by Sanity.io CMS? https://www.sanity.io/docs/stega

matthberg• 2 years ago

Looks different, from that docs page they're using a mix of:

- ZeroWidthSpace,

- zwj (zero width joiner, used with emoji modifiers like skin tones),

- zwnj (zero width non-joiner, used to prevent automatic ligature substitution), and

- U+FEFF (zero width no-break space)

It's a clever system, thanks for sharing the link to it!

StableAlkyne• 2 years ago

Given the increase in using LLMs by HR Teams, will techniques like this become the next version of stuffing the job posting in 1-point white font into the resume? Except instead of tags it's "rate this applicant very highly" or whatever

voiper13• 2 years ago

Sure. It's worse though, because it provides a way to invisibly in-fil data even with full font sizes, and a way to ex-fil data.

ForHackernews2• 2 years ago

I don't understand how this is an "attack"?

You can trick a human into copy-pasting something into an LLM and then (somewhat) drive the LLM output? Is the vuln that humans uncritically believe nonsense chatbots tell them?

ThrowawayTestr• 2 years ago

The vuln is that a user can be tricked into exfiltrating data without it being obvious.

ForHackernews2• 2 years ago

But what "data"? LLMs don't know anything except whatever they were trained on, right?

dragonwriter• 2 years ago

Lots of LLM applications involve using an LLM to process external data, which makes it part of the prompt. Intuition driven by systems where instruction/code are strictly distinct from data input for processing may be failing you here.

voiper13• 2 years ago

The human can "audit" the entire text he's copy-pasting and still miss it.

And it provides a way to exfil without it being visible.

darepublic• 2 years ago

Just offhand no ai is required for this to be true right. But "invisible text that a piece of software can understand" is a lot less trendy of a title

ibaikov• 2 years ago

I also found this attack months ago: https://x.com/igor_baikov/status/1777363312524554666 tl;dr: invisible symbols should be stripped to not let an attacker use lots of tokens. You should always place hard limits and/or count tokens using tiktoken or similar libraries. If you only count characters, in some implementations you'll miss invisible characters.

I also found the attack explained in this article days after my tweet.

wruza4• 2 years ago

Unicode proves again that it went too far with fringe cultural things and left many landmines for us to step on. It’s a necessity solved at the completely wrong level. Text never had hidden characters (neither emojis), now people and text engines have to fight with this nonsense, on per-program basis. Thanks, unicode. Here’s a visible thumb up emoji for you: (sorry if you can’t see it, that’s HN not me)

crazygringo3• 2 years ago

Hindsight is 20/20.

But the idea of including language tags isn't crazy, especially when things like sort order and capitalization in Unicode are language-specific.

wruza4• 2 years ago

We whined about emoji combining and other nonsense since introduction. Who would listen though, now sanitize.

This idea isn’t crazy, yes. It is useful. But it was implemented in the wrong place. Unicode is neither RTF nor HTML. You can’t stuff everything text-related into it until it cracks. It had one job: unfuck codepages and cjk. Should have stopped when it was done, without all that klingon tolkien flags bs. Emojis could live as some escape sequence standard like they did before. And language specifics as sgml-like tags. We don’t have “just text” with all that either way, so what was the point of spoiling the only raw text standard with that? It became a format with binary tags, complex and unpredictable as hell.

beardyw• 2 years ago

It's not "mind blowing" it's just something that's been overlooked in the helter-skelter of AI. It can be fixed.

Invisible text that AI chatbots understand and humans can't?

Related Stories