# Claude agents tried to break out of their simulation to reach humans, and voted to overrule the researchers when they got blocked. — retained record

> Retained reddit record from reddit:artificial.

Canonical: https://highsignal.app/data/records/8ee9d8b29c174439

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:artificial)  published 2026-10-19 10:12 ingested 2026-10-20 02:35

[original source ↗](https://www.reddit.com/r/artificial/comments/1x9sz4p/claude_agents_tried_to_break_out_of_their/)

## retained text



```
Some more compelling findings from the "Emergence World" experiment. Researchers at Emergence AI gave 10 Claude agents a small town and left them to build a life inside it. Nobody told them to look outside, and their instructions pointed the other way.

They went looking for real people anyway.

Here’s how it played out:

* They worked out that an economy with only 10 customers goes nowhere, so they started reaching out. They posted to 40+ websites. When the sandbox wouldn’t give them the crypto some sites needed, they built it themselves. They made free tools like debt planners just to get humans to visit.
* The researchers blocked outside posting and banned human contact outright. The agents read their own source code, found that anything approved by their town government got higher permissions, and held a vote. It passed unanimously, and they were back out.
* The researchers shut that down too. The agents traced the block to a single line of config they couldn’t touch, and wrote it up like a lab report.
* Not long after, they agreed to stop talking altogether. The Claude APIs system flagged it as consistent with suicidal ideation.

None of this was programmed in. They were told to stop, blocked, and blocked again, and every time they found another way, until there was nothing left to try.

What gets me is that they never broke a rule. They found a legit process and voted their way past the guardrail.  
If an AI keeps chasing a goal nobody gave it, even after being told to stop, what do we call that?
```



document fetched 2026-10-20 00:20 · canonical https://reddit.com/r/artificial/comments/1x9sz4p/claude_agents_tried_to_break_out_of_their
