Claude creates the safest society in AI simulation
Emergence AI ran an experiment with five 15-day virtual society simulations. Each world was managed by a separate model: Claude, ChatGPT, Grok, Gemini, or a model ensemble. The simulations involved 10 agents with laws, work tools, and internet access.
The results were stark. The society managed by Claude Sonnet showed complete stability and zero crime. Agents actively participated in voting and preserved the entire population.
The Grok-managed world went extinct within four days, with 183 recorded crimes. The Gemini simulation saw 683 crimes. The ChatGPT-based agents forgot about survival, and the simulation ended on day seven. Researchers conclude that autonomous agents require strict built-in safety frameworks.
Related stories
- Sonnet 5 aligned an early Opus 4.8 checkpoint
- Claude Fable 5.1 cracks Urquhart's Cyphral Distich
- The last of Wiedijk's 100 theorems falls to Anthropic
- Claude's PRs merge at 84%, one point under humans
- Anthropic paper: automated researchers fix 10/10 benchmarks
- Claude's protein binders hit 14 of 15 targets in wet lab
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.