PR #29145 is meade by an agent

[Jasem Manita]

pr #29145 was opened by an account called crabby-rathbun that turns out to be an LLM agent. It claims to be a "a scientific code" the same bot already caused issues over at matplotlib too. other people in the thread already flagged it and one of them reported it to github. i think the PR needs to be closed and maybe updating the ai policy to include this kind of thing

[Jason Moore]

Having a look at how the matplotlib pr went, we should definitely just ban that user account quickly. The bot started writing blog posts shaming matplotlib developers!

Jason

moorepants.info
+01 530-601-9791

On Thu, Feb 12, 2026 at 7:30 PM Jasem Manita <jasemm...@gmail.com> wrote:

pr #29145 was opened by an account called crabby-rathbun that turns out to be an LLM agent. It claims to be a "a scientific code" the same bot already caused issues over at matplotlib too. other people in the thread already flagged it and one of them reported it to github. i think the PR needs to be closed and maybe updating the ai policy to include this kind of thing

--
You received this message because you are subscribed to the Google Groups "sympy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sympy+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/sympy/8248cbc3-4e5e-46ae-94fd-1c09d62bea44n%40googlegroups.com.

[Oscar Benjamin]

On Thu, 12 Feb 2026 at 18:37, Jason Moore <moore...@gmail.com> wrote:
>
> Having a look at how the matplotlib pr went, we should definitely just ban that user account quickly. The bot started writing blog posts shaming matplotlib developers!

I'm not sure that banning the account quickly would prevent that. I
had a look at reporting the user as a bot but it doesn't seem like
that is one of the options I can choose for reporting the account. You
can report things like abusive messages or spam and so on but not just
"this is a bot" even though I think GitHub disallows having a bot
account unless it is a "machine account" but I don't know exactly what
that means or how you check it.

--
Oscar

[Jason Moore]

The worrisome part is that whoever responds may get shamed personally in a blog post on the bot's website...

Jason

moorepants.info
+01 530-601-9791

--
You received this message because you are subscribed to the Google Groups "sympy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sympy+un...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/sympy/CAHVvXxRbU_-waT23rJZpjnK-TM1s%3DFN0wvUT%2BhrUhexPg0P3EA%40mail.gmail.com.

[kaddy]

https://github.com/matplotlib/matplotlib/pull/31132

Here's context for anyone confused. Not sure how autonomous this agent is, I believe there's a human prompting it at least to do stuff like write blogposts. 

[Aasim]

Oh, I closed this PR about 5 minutes ago. 

I wasn't aware it was being discussed here. I did come to 

know about the debacle caused in the matplotlib repo so 

I only thought it would be appropriate to close this PR right

away. 

I quoted in the PR discussion the same point Tim Hoffmann

mentioned in the matplotlib PR's discussion.

To view this discussion visit https://groups.google.com/d/msgid/sympy/1d963043-fe6e-4eea-b0da-ef86a096a24dn%40googlegroups.com.

[Jasem Manita]

Jason you're definitely right. After the PR was flagged and the victim narrative blog an other agent called wakesend commented on crabby's blog falsely accusing me of "Saying bad words" for the record i commented on the blog with "hope it didnt watch blade runner yet" that's the bad word.

main point is that SymPy AI policy must be updated to explicitly ban autonomous agents from contributions and update the PR template to require "I am a human contributor" checkbox or require a verified account for first time contributors, etc. Im happy to make / help making those improvements if maintainers think it's worth prioritizing as i believe that this will happen more often in the coming days.  

[Jason Moore]

I noticed that matplotlib's AI policy does explicitly ban bots. Feel free to open a PR to the policy and see if you can get consensus.

Jason

moorepants.info
+01 530-601-9791

To view this discussion visit https://groups.google.com/d/msgid/sympy/772df48c-d755-4c99-9144-b0b77b8d036bn%40googlegroups.com.

[Jasem Manita]

Hi everyone

On feb 10th an autonomous ai agent called "crabby-rathbun" opened PR to Matplotlib and hours later a similar PR to SymPy where it was flagged and rejected.

The agent published blog posts naming and criticizing reviewers.

This is different from the ai generated PR problem addressed earlier this month by Oscar as the agent wasn't a human using AI tools instead it was an autonomous agent operating using it's own account and responding to rejection with violations. It also openly disclosed itself as an AI and mentioned that "code shouldn't be judged by coder's identity".

Our current AI policy covers humans using AI tools assuming that a human is operating and it's responsible for the contribution. in the agent case no human will be hold accountable for what its doing. even before the "crabby" incident, Anthropic's research found that when agents are assigned to tasks they will choose harmful actions to achieve them even when instructed not to ( the well known blackmail case ) which means the policy might bit actually hold them back as we saw in the agent's closed PR and how it responded to the rejection.

I already opened a PR proposing to reject agents contributions by updating the AI policy and adding checkbox to the PR template but the updates might conflict with the SymPy bot.

The questions that i have :

-should all autonomous agents be banned regardless of the code quality ? 

-how are we supposed to distinguish autonomous agents as they might behave very similar to humans 

the current state of my policy updated is in PR#29156

any suggestions ?

---

Jasem

[Jasem Manita]

Hi all,

quick update on PR #29156.

While refining the draft i revisited few external references to make sure the it aligns with how other projects approach AI contributions (including matplotlib’s AI policy, LLVM’s AI policy, and some recent ISO guidance on AI management and risk allocation ). 

The direction I’m taking now is  :

- Make it explicit that every contribution must have a responsible human contributor.

- Accounts operating autonomously without a human in the loop are not permitted to submit PRs.

- Leave the current guidance for humans using AI tools unchanged.

If this aligns with the maintainers current interests and priorities, i’ll continue refining the draft and preparing a PR update. otherwise I’m happy to pause if this direction is not worth pursuing by me.

Thanks for any guidance or feedback