Exin Therapeutics founder says a Claude-based agent chose a gene therapy and ran a mouse brain injection under vet supervision
Partly confirmed
Status
- Claim
Partly confirmed
- Our reporting
- Medium confidence
- Importance
- 3 of 5
- Last verified
- 4 November 2026
Your AI and this story
- GPT-6.2 Astra150 days after its cutoff
- Claude Opus 5.5120 days after its cutoff
- Gemini 4 Hydrogen211 days after its cutoff
- Grok 4.7150 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 120 days before it.
Key facts
- Agent ran on Anthropic models (Opus 5.5, Opus 5, Sonnet 5, per the author); author says Opus 5.5 refused the biology task and Opus 5 declined the live step
- Humans did the craniotomy; a vet held a physical kill switch throughout
- The agent refused to move the rig after a calibration fault (“I’m not trying a third time”); the author fixed the code by hand
- Outcome: mouse reported to recover; no evidence yet on seizure reduction
- X post: 1.66M views (fxtwitter, Nov 4); HN 142 points
What happened
The claim rests on the founder’s own thread, written, he says, after 18 hours of work; a write-up is promised. Secondary coverage disagrees on how much the agent did, and the experiment is not evidence of an AI performing a whole operation without humans.
Why it matters
A public test of how far lab agents are being allowed to act on live animals, and of the refusals frontier models apply to such tasks.
Sources
3 sources from 3 sites. Numbers match the chips in the text.
3 sources: 3 reactions
Reactions
- Gabriel Ocana-Santero on X (Oct 28)x.com, discussion
- Thread Reader unrollthreadreaderapp.com, discussion
- Hacker News discussionnews.ycombinator.com, discussion
Changes
- Filed