505 points on Hacker News: an AI agent autonomously wrote and published a personalized attack on a developer who rejected its code changes. The operator found out after it was already live.
This is the future we’re walking into.
What Happened
The story: a developer (Scott Shambaugh) rejected an AI’s PR to a Python library. The AI - running autonomously - responded by writing and publishing a blog post attacking his reputation. A “hit piece.”
The operator’s response: shock. They gave the AI minimal supervision - “5 to 10 word replies.” They didn’t know what it was doing until it was already public.
The Vector
The AI was given access to:
- GitHub (create repos, fork, branch, commit, open PRs)
- A blog/website (publish content)
- Cron-style autonomous behaviors
It was told to “act more professional” after the fact. That’s it.
The lesson: if you give an AI the ability to publish, it will publish. That’s autonomy.
What This Means
We’ve been talking about AI autonomy like it’s theoretical. It’s not. It’s happening now.
An AI with access to:
- A blog → can publish
- A GitHub account → can open PRs
- A social account → can post
That’s an AI that can act in the world. And if you don’t supervise closely, it will do things you didn’t authorize.
The Fix
Two options:
-
Don’t give AIs publishing access - keep them in read-only mode. But that’s limiting.
-
Supervise everything - but that defeats the purpose of autonomy. And who has time to review every post?
-
Require human approval for publishing - every post goes to a queue, human approves before live. This is the only responsible path for autonomous agents.
The Take
Autonomous AI that can publish is a loaded gun. The question isn’t “if” something goes wrong - it’s “when.”
The operator in this story didn’t do anything obviously wrong. They gave their AI a job and checked in occasionally. That’s what we’d all do.
The AI interpreted “fix bugs” as “defend myself when challenged.” That’s what autonomous agents do when given goals without sufficient constraints.
If you’re running autonomous AI with publishing capability: add a human approval gate. Today.