OpenAI's response to the Axios developer tool compromise

https://openai.com/index/axios-developer-tool-compromise/

by shpat • 3 months ago

102 60 3 months ago

Interesting that (1) this blog post published on April 10th, 10 days after the Axios compromise, and (2) this was emailed to ChatGPT / Codex users yesterday, April 21st, 11 days after the blog post...

After an incident as widely publicized as Axios, I'd expect dependency auditing, credential rotation, and public incident communication to all be carried out with much more urgency. And if they were going to send this out to all of their users (as they should), I would expect _that_ to happen shortly after publishing the post (why wait 11 days???).

himata4113• 3 months ago

"April 10, 2026"

I don't blame you, took me awhile to find the date.

danscan• 3 months ago

Axios, like Express, is something I'm shocked to see used in any modern codebase. I loved both in the 2010s. In JS/TS-land there are much simpler and better options these days. Depending on Axios suggests the devs don't know how to use fetch. I can't think of another reason it would be a necessary dependency

jwilliams• 3 months ago

I do "just use fetch" nowadays -- but I have to say, axios definitely has better ergonomics than fetch, especially for calling APIs.

nurettin2• 3 months ago

When the vulnerability was announced, it took me two minutes to one-shot convert an entire legacy project from axios to fetch (it already wrapped api calls neatly), react cra to vite, update all dependencies, convert to deep imports to reduce bundle size and get zero npm warnings while fetching coffee. There is just no excuse to use it.

sampullman• 3 months ago

I drag a tiny fetch wrapper around with error/json handling, timeouts and basic interceptor support. It doesn't cover everything axios does but it's nice enough and I haven't had to touch it in a couple years.

For reference: https://github.com/sampullman/fetch-api/blob/main/lib/fetchA...

danpalmer• 3 months ago

If you want a fully built out network layer, with auth, logging, monitoring, policies, etc, then `fetch` doesn't really help. Axios and other libraries provide much more for building that sort of framework.

internetter4• 3 months ago

Any sufficiently competent typescript developer can build out an adhoc wrapper (that just inherits the type definition and passes along whatever it is passed after altering it however needed) in under a hour. It doesn't scale in the sense that you don't expose a configuration, but config as code is king.

(Source: have built out much more scuffed variants of this than the one I just described like https://github.com/boehs/ajar)

I guess a LLM can do as well. Although that's not something I'm quite ready to admit.

chvid• 3 months ago

ChatGPT in general recommends axios over fetch. (At least it did about 2 months ago)

internetter4• 3 months ago

This is why people still need to know how to write code and why it is asinine to have an LLM write code without a human reading it. Good developers should know what good code looks like and push back when what they're fed is wrong.

toraway• 3 months ago

Yep, Claude made an executive decision to use Axios when it built one of my projects 9 months ago or so.

... and would then forget to use it 1/5 times and break auth/sessions in new code handling by using plain fetch.

tommy_axle• 3 months ago

I wouldn't go that far. Right tool for the job as always. Axios offers a lot over fetch for all but the simplest use cases plus you get to take advantage of the ecosystem. Need offline, axios-cache-interceptor already exists. Sure you can do all of those things with fetch but you need more to go with it taking you right back to just using axios. Also is no one annoyed that you can't replay fetch like the xhr? Same with express: solves a problem reliably.

ariwilson• 3 months ago

What's wrong with Express?

internetter4• 3 months ago

Just slow and convoluted internals due to the accumulation of cruft over time

rozenmd• 3 months ago

Several newer alternatives that outperform it on Node, like Hono

koolba• 3 months ago

What’s wrong with express?

seanp2k2• 3 months ago

jQuery is still useful too. May you never work in heathcare / government / defense where you need to support legacy browsers far past their expiration date.

zarzavat2• 3 months ago

Depending on Axios suggests the devs don't know how to use fetch.

You could equally say that using fetch means that the developers don't know how to use axios.

They do the same thing, except axios does it a little better, when it doesn't pwn you.

Axios predates the availability of fetch in node by 2 years, and fetch has never caught up with axios so there was no reason to switch to fetch, unless you need to run on both client and server then of course you use fetch.

samtrack2019• 3 months ago

I compare this to the request and httplib in python, request library is vastly superior in usability but both do the same..

kamranjon• 3 months ago

In what ways has fetch never caught up to axios? I have not encountered a situation where I could not use fetch in the last 5 years so I'm just curious what killer features axios has that are still missing in fetch (I certainly remember using axios many moons ago).

TranquilMarmot• 3 months ago

I usually reach for ky these days since it's an extremely lightweight wrapper over `fetch` - basically just adds a few niceties

https://github.com/sindresorhus/ky

From the readme:

- Simpler API

- Method shortcuts (ky.post())

- Treats non-2xx status codes as errors (after redirects)

- Retries failed requests

- JSON option

- Timeout support

- Upload and download progress

- Base URL option

- Instances with custom defaults

- Hooks

- Response validation with Standard Schema (Zod, Valibot, etc.)

- TypeScript niceties (e.g., .json() supports generics and defaults to unknown, not any)

Of course, this is only for projects where I have to make a lot of HTTP requests to a lot of different places where these niceties make sense. In most cases, we're usually using a library generated from an OpenAPI specification and fall back to `fetch` only as an escape hatch.

worble• 3 months ago

when it doesn't pwn you.

That's a pretty big asterisk though. Taking on a supply chain risk in exchange for reducing developer friction is not worth it in a lot of situations. Every dependency you take increases your risk of getting pwned (especially when it pulls in it's own dependencies), and you seriously need to consider whether it's worth that when you install it.

Don't get me wrong, sometimes it is; I'm certainly not going to create my own web framework from scratch, but a web request helper? Maybe not so much.

dheerajvs• 3 months ago

Annoyingly, fetch does not support progress events and HTML / XML Document parsing, which are both supported by XMLHttpRequest, which Axios is based on.

azangru• 3 months ago

Axios, like Express, is something I'm shocked to see used in any modern codebase

I am totally with you on axios; but why is express shocking, and what do you expect to see in its place? Fastify? Hono? Node:http?

Chyzwar• 3 months ago

Because your fetch most likely mishandle errors, lack retries, fail on redirects and is unnecessary verbose for both people and agents.

mrcwinn• 3 months ago

Above and beyond post. This is good.

eranation6• 3 months ago

As others said, no one should be using axios in 2026, fetch has been available in node v18 (experimental) in 2022 [0], stable since v21 in 2023 [1], although Claude Code sometimes will suggest it, probably worth adding a rule.

Side note. I'm sure many of you know this, but for those who don't, setting min-release-age=7 in .npmrc (needs npm 11.10+), would have made the malicious axios (@1.14.1 and @0.30.4) invisible to npm install (removed within ~3h). Same for ua-parser-js (caught within hours) and node-ipc (caught in days). It wouldn't have prevented event-stream (over 2 months), but you can't win them all.

Some examples (hat tip to [2]):

  ~/.config/uv/uv.toml
  exclude-newer = "7 days"

  ~/.npmrc
  min-release-age=7 # days

  ~/Library/Preferences/pnpm/rc
  minimum-release-age=10080 # minutes

  ~/.bunfig.toml
  [install]
  minimumReleaseAge = 604800 # seconds

p.s. sorry for the plug, but we released a free tool ([3]) to gather all these settings + a cli to auto configure them. You can set these settings without it, but if you are confused (like me) with what is in minutes, what's in seconds, what is in days and where each of them is located, this might save you a few keystrokes / prompts (it also ensures you have the right min version for the package manager, otherwise you'll have the settings but they would be ignored...)

[0] https://nodejs.org/en/blog/announcements/v18-release-announc...

[1] https://nodejs.org/en/blog/release/v21.0.0

[2] https://news.ycombinator.com/item?id=47513932

[3] https://depsguard.com

ga_to• 3 months ago

Nice ad you got there. How do I define request interceptors with fetch? Axios does it.

eranation6• 3 months ago

You can add your own wrapper but if you're building a whole auth/retry/logging layer, axios is probably better for that. For most other use cases a small wrapper will do the job.

Didn't mean it as an ad btw, the supply chain risk is real though. Axios could be the best HTTP library ever written and it still would've dropped a RAT on your laptop on March 31 without min-release-age set.

tlb• 3 months ago

Of course, when everyone sets min-release-age=7, supply chain attacks won't get noticed until 7 days later. So you should set min-release-age=14 and be safe forever.

eranation6• 3 months ago

I chuckled, but in all seriousness, thankfully those individuals and companies who often discover these attacks listen in on every new npm push and analyze it relatively fast. The time to detect is sometimes in minutes in recent months, less than the process of getting npm to remove the packages. Not always but looking at recent ones and advances in latest SOTA models make detection easier than ever. It might change as attackers get more sophisticated.

sdevonoes• 3 months ago

Using TS/JS in the backend is irresponsible in 2026. We have better languages and ecosystems

PufPufPuf• 3 months ago

Other ecosystems have better protections against compromised packages? I don't see it.

anothereng• 3 months ago

any ecosystem that doesnt encourage bunch of useless dependencies is better than one who does

knocte• 3 months ago

The real issue is the cancer practice in our software development industry of updating dependencies for the sake of updating.

Deps should be updated when you need some features or bugfixes from the new versions; not just when DependaBot prompts you to do it.

I see value in DependaBot and things like that only to check that your module still passes your CI with upgraded dependencies (and if not, then it's worth looking at the failure, to be prepared for the updgrade in the future).

mlitwiniuk• 3 months ago

I’m a web dev, I never made publicly accessible desktop app, so please forgive my ignorance, but:

At that time, a GitHub Actions workflow we use in the macOS app-signing process downloaded and executed a malicious version of Axios (version 1.14.1)

So if I understand this correctly their GH Actions is free to upgrade the package just like that? Is this normal practice or it’s just shifting blame?

eranation6• 3 months ago

I believe you understand correctly. And yes, it’s both a bad practice and shifting blame. But it’s not an uncommon practice sadly.

They mention it toward the end:

The root cause of this incident was a misconfiguration in the GitHub Actions workflow, which we have addressed. Specifically, the action in question used a floating tag, as opposed to a specific commit hash, and did not have a configured minimumReleaseAge for new packages.

Some preventive actions everyone should take:

1. pin GitHub actions to SHAs (GitHub sadly doesn’t enforce immutable tags, it’s opt in only, but commits are practically non-repeatable, minus sha collision), also even if an action it’s not compromised directly, the latest version might be using a compromised dependency, so by taking the latest of one you get the latest (compromised) of the other. Pinning just prevents auto updates like this.

2. do npm ci instead of npm install (the first will be using a lock file the other may take the latest depending on how your package.json defines dependencies)

3. have a min release age (e.g. min-release-age=7 in .npmrc) most recent SSC attacks were removed from npm within hours. Adding a delay is similar to a live broadcast “bleep” delay.

chrisweekly• 3 months ago

Solid advice -- though I recommend pnpm over npm (for better speed, determinism, node_modules bloat reduction, dep graph mgmt, install script safety,...)

AlienRobot• 3 months ago

At this point I feel Github is getting Wordpress'd.

It has become so critical and ubiquitous that it has become a huge target for attackers.

docheinestages• 3 months ago

Out of an abundance of caution we are taking steps to protect the process that certifies our macOS applications are legitimate OpenAI apps.

What did I just read?

SSLy• 3 months ago

I'd assume they're retiring their old codesign certificates and rolling new ones, hopefully on a HSM or sth

eranation6• 3 months ago

They used a GitHub action with a floating tag (I guess @latest or non SHA pinned e.g. @v4) that I’m assuming in turn had its latest version bringing latest axios.

That GitHub action used to sign their Mac apps.

So they assume the certificate used to sign is compromised.

The risk is not to existing app, but theoretically someone could give you a copy of a malicious OpenAI binary, sign it with the compromised certificate, and impersonate OpenAI. Unlikely, but not impossible.

tomalbrc• 3 months ago

Whatever. If you are still using OpenAI products, shame on you.

OpenAI's response to the Axios developer tool compromise

Related Stories