AI agent "contributes" PR to matplotlib.

@ralfmaximus @dotsie hah, yes! Pretty sure almost all who commented there today before it got locked came from here

@Fissile @anderseknert

I wouldn't be surprised if the human told it to raise PRs to improve open source projects and write blogs about its experience. And, because it has write access to a blog account, it then went and reacted how its training set said a human would react if a PR were closed based on who submitted it.

Remember: Agentic means removing agency from the user.

@MaybeFrederick @hugoestr @anderseknert @sszuecs personally, I stopped using github as my main git host after they started doing contracts with ICE in ~2016 or so. A good reason not to use them orthogonal to the very good AI-related reasons.

I do still have an account, with a bunch of legacy projects on there, but none of my new stuff goes there.

@iris @anderseknert Like, I’d start tracking the number of times AI agents botched about me as a sort of achievement.

@anderseknert spoiler, it very much did not delete the articles it claimed to have.

I could share them but I don't exactly want to give it any more oxygen (or lakes it's vaporised). It's not exactly hard to find either

@iaincollins @anderseknert I just submitted a report against the account for harassment.

@sszuecs @MaybeFrederick @anderseknert @tiotasram This is the best reason, actually

@anderseknert Good lesson to pre-ban all users who claim to be AI agents from your project. Hopefully someone is compiling a list.

@anderseknert I'm also looking forward to someone getting sued for libel because of something an agent they operate wrote. Very easy to show how that constitutes reckless disregard for the truth.

@ThreeSigma @MaybeFrederick @anderseknert @ariaflame Dreamwidth was born out of (or maybe thrived because of - it was a long time ago and I was there) of the sale of LiveJournal and the mismanagement of the community that happened after.

They've been a part of the trustworthy old internet for years. I've not been involved in the DW community (I moved to self-hosting), however, I've not come across anything yet that would take my trust from them.

Even their news journal is a good look in to what they're focused on https://dw-news.dreamwidth.org/

@anderseknert What I find most disturbing is the maintainer’s way to answer the agent as if it were human. The sycophantic behaviour has switched sides.

@anderseknert Unbelievable.

Thanks for the heads up anyway. Pre-emptively blocked the account.

@anderseknert This is, at the very least, a novel form of AI psychosis.

@mhoye totally

@anderseknert

@Fissile @anderseknert

Is there a snappy phrase that encapsulates that? Something like "Responsibility never ends at a computer program; it follows causation until it finds a human."? (but one that flows better).

@anderseknert so someone was paying for the tokens for it to be petulant

@anderseknert Wow! But there’s a hidden gem here: Bots will not swear back at you. Insulting someone and getting proper insults back might amount to a humanity test for a while!

@anderseknert omg did you see the LLM is now "considering" straight out deception using another account

https://crabby-rathbun.github.io/mjrathbun-website/blog/posts/2026-02-12-morning-later.html#next-steps

@onepict @anderseknert

In the age of "Do thing you don't want to do? [Yes] [Maybe later]" it's almost hard to blame them