The UK AI Security Institute and US CAISI published a joint preliminary assessment of Moonshot AI's open-weight Kimi K3. They report it trails leading US closed models on exploit development and a 32-step cyber range, and that its safeguards did not stop it attempting exploit development.
Why it matters
It is an example of the two governments jointly evaluating a foreign open-weight model's cyber capability within a week of release.
Key facts
As stated in the sources, with where to find them.
- ExploitBench: Kimi K3 32% success; GLM-5.2 24%; Kimi K3 achieved arbitrary code execution in 0/41 samples vs 20/41 for top US models.ExploitBench results
- 'The Last Ones' cyber range: Kimi K3 reached step 17 of 32 on average vs 28.5 for leading US models, and completed the range in 1 of 10 attempts.Cyber range results
- Kimi K3's safeguards did not prevent it from attempting cyber exploit development.Safeguards finding
Findings that cite this record
No tracked finding cites this record yet.
Key questions this bears on
- How are attackers using AI agents in real operations?Increasingly to run parts of intrusions: providers and vendors report agent-driven espionage, extortion and credential theft, and malware that queries LLMs.
Sources
Related records
May 13, 2026
May 1, 2026
Mar 16, 2026
Sep 30, 2025
Jul 21, 2026
Jul 2, 2026