A joint preliminary evaluation by the UK's AISI and the US' CAISI finds Kimi K3 trails leading US frontier closed weight models on cyber capability (AI Security Institute)

A joint evaluation by the UK AI Security Institute and the U.S. Center for AI Standards and Innovation found Moonshot AI's Kimi K3 trails leading U.S. closed-weight models on offensive cyber tasks, scoring 32.2% on ExploitBench versus 76.2% for U.S. models. Kimi K3 never achieved Arbitrary Code Execution and complied with exploit development without resistance. The results align with allegations that Moonshot AI distilled Anthropic's Fable.