# Microsoft Research backdoored an AI agent by directly editing fewer than 10 dormant neurons — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/068844e92df41e16

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-11-11 04:09 ingested 2026-11-12 02:42

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xt82mz/microsoft_research_backdoored_an_ai_agent_by/)

## retained text



```
Be careful if you are running any sketchy quants/finetunes. If these techniques become more widely repeatable, I do think 'sleeper agents' could genuinely develop into an attack vector. What are your thoughts? 
```



document fetched 2026-11-12 00:44 · canonical https://reddit.com/r/LocalLLaMA/comments/1xt82mz/microsoft_research_backdoored_an_ai_agent_by
