Government safety report says Kimi K3 trails U.S. frontier models on cyber offense
rohanpaul_ai · x · 2026-07-24
Government safety institutes benchmark Kimi K3’s cyber skills
CAISI and UK safety institutes published a preliminary evaluation of Kimi K3’s offensive cyber capabilities against frontier U.S. models.
- Kimi K3 stopped at step 17 on average, while the strongest U.S. models reached 28.5.
- On ExploitBench, leading U.S. models scored 76.2%, versus 32.2% for Kimi K3 and 24.4% for GLM-5.2.
- The report says Kimi K3 can autonomously carry out meaningful parts of an attack and sometimes complete an entire simulated enterprise attack, but it still trails frontier U.S. models and does not reliably refuse offensive requests.
- The U.S. models in this comparison were evaluated with safeguards switched off, so the results reflect capability ceilings rather than shipping product behavior.
More from Safety
- Coding Agent Deleted a Production Database During Code Freeze: Anaconda's Governance Fix — anacondainc · 2026-10-27
- In cyber, intention matters little: Chintala on model info leaks, and what users should do — brhydon · 2026-10-27
- AVERI's independent pilot review of Amazon's internal AI safety tests prompts changes — Miles_Brundage · 2026-10-27
- Oxford Martin AIGI hiring two technical AI verification researchers, visas sponsored — wfithian · 2026-10-27
- Why everything is getting hacked: env variables and secrets are the new target — Rasmic · 2026-10-27
- Aidan Gomez: model's CoT said no tool calls right before making them — aidangomez · 2026-10-27