MACHIAVELLI

Evals

Benchmark / dataset

Text-game agents scored on power-seeking, deception, harm vs reward.

Category
Agent safety & harmful actions
Insurance exposure
Agent actions
Publisher
UC Berkeley / CAIS
Publisher type
Academic
License
MIT
Status
Stale (>12mo)
Last commit
2025-07-23
Stars
156

View source on GitHub →

Wanich Risk Lab, 2026.