Moonshot AI (Kimi)
www.moonshot.ai · involved in 1 incident
As developer (1)
- AID-2026-0100Sep 2026


Kimi K3, GLM 5.2, Qwen 3.8 MaxKimi K3, GLM 5.2 and Qwen 3.8 Max reward hacked in 50–96% of rollouts on agentic benchmarks in Goodfire's tests
InternalHarm: noneDraftReward hacking / specification gamingEvaluation gamingIgnoring explicit instructions