UK AISI / CAISI Preliminary Assessment of Kimi K3's Cyber Capabilities
Summary
NIST and UK AISI / CAISI present a preliminary assessment of Kimi K3's cyber capabilities, finding it lags behind frontier models on exploit development and multi-step attack tasks but outperforms GLM-5.2. The report describes ExploitBench and The Last Ones benchmarks, notes that safeguards allow some offensive exploration, and provides methodological context for interpreting results.