UK and U.S. Institutes Assess Kimi K3's Cyber Capabilities with Mixed Results
The UK Artificial Intelligence Security Institute (UK AISI) and the U.S. Center for AI Standards and Innovation (CAISI) conducted a joint evaluation of Moonshot AI's latest model, Kimi K3. Released on July 16, 2026, Kimi K3 was assessed for its cyber capabilities, particularly in exploit development. The evaluation used ExploitBench, a benchmark focused on exploit development, to measure the model's ability to progress along the software exploitation ladder. Kimi K3 showed a higher success rate in some areas compared to other models but failed to achieve arbitrary code execution, a critical milestone in exploit development. The model's performance was also tested on 'The Last Ones' cyber range, where it performed below leading U.S. models, completing only 17 out of 32 steps on average. 
