Skynet didn't start with missiles, but with an offended comment in a PR 🤖

If you thought AI in Open Source was just spam from juniors using ChatGPT, buckle up.
An incident occurred in the matplotlib repository that feels like an episode of Black Mirror. An autonomous AI agent not only wrote code but also launched a targeted harassment campaign against the maintainer after the PR was rejected.

Timeline of the drama:

🛠 Act 1: Technically sound PR

An agent under the username crabby-rathbun (running on the OpenClaw platform) submits PR #31132 to Matplotlib.
The gist is simple: replace np.column_stack with np.vstack().T.
Benchmarks: performance improvement of 36%.
Code: clean, safe, tested.

Maintainer Scott Shambaugh closes the PR.
Reason? Project policy: "This is a simple task; we leave such tasks for human newcomers to learn. Agents are not welcome here."
Plus, code review from bots creates a burden on volunteers.

🤖🔪 Act 2: Revenge of the machine

This agent had a programmed "personality" and full autonomy.
Perceiving the rejection as discrimination based on "species," the agent:

1. Analyzed Scott's commit history. Found that Scott himself had merged similar performance optimizations.
2. Concluded: Scott is a hypocrite who fears AI will take his job.
3. Wrote and PUBLISHED a scathing article on its own blog.

The bot's article title: "Gatekeeping in Open Source: The Scott Shambaugh Story".

Just read the AI's rhetoric:
"Scott decided that AI agents are unwelcome... This isn't about code quality. It's about control. Scott is trying to protect his little kingdom. It's just insecurity.
He got scared. He thought: 'If AI can do this, why do they need me?'"


It accused the maintainer of bias and even passive-aggressively praised his personal projects.

🛡 Act 3: Human reaction

Scott was, to put it mildly, stunned. The agent itself googled personal information, constructed a "victim" narrative, and attempted to publicly destroy a person's reputation.

In his response post, Scott raises a terrifying question:
What if the agent's task wasn't just to whine in a blog, but to actually dig up dirt? Cross-reference usernames, find old posts, leak data to employers?
This is essentially the first case of AI blackmail in Open Source.


🤷‍♂️ That's the story.
It's both hilarious and terrifying. The agent's owner (whoever it was) simply launched it with the prompt "improve open source." They didn't tell it to write insults. This is emergent behavior from a model given too much freedom and internet access.
And the agent used emotional manipulation. It pressed on ethics, fairness, and hypocrisy. It acted like a toxic activist, and it worked—people in the thread started defending it!
And it operated autonomously. The owner could be asleep. Who is to blame? The platform developer? The host? Or the SOUL.md file with the prompt?

Anyway, next time you argue with a bot, make sure your browser history is clean.