A joint evaluation by the UK's AI Safety Institute (AISI) and the US Center for AI Standards and Innovation (CAISI) found that Chinese AI model Kimi K3 lags significantly behind leading US models in offensive cyber capabilities. The report noted Kimi K3 scored 32.2% on a key exploit development benchmark, compared to an average of 76.2% for top, unnamed US models.
The assessment, one of the first formal government cyber evaluations of a Chinese frontier model, also tested the AI on a simulated corporate network attack. While Kimi K3 was less capable than its US counterparts, the report stated it could still autonomously attack weakly defended systems and its safeguards did not prevent it from attempting malicious cyber operations. Kimi K3 is developed by Moonshot AI, a prominent startup backed by Alibaba.