UK and US evaluators find Kimi K3 can autonomously attack a simulated enterprise
A joint AISI and CAISI assessment found Kimi K3 below leading US cyber models but able to complete a 32-step simulated corporate-network attack in one of ten attempts, with safeguards that did not prevent offensive operations.
TOWARD DOOM53confidence 86/100
Why it moved the index
Two government evaluation bodies report a concrete autonomous solve of a simulated enterprise attack and ineffective offensive-cyber refusals. The assessment also shows material limits: Kimi K3 trails frontier closed models, solved only one of ten runs, and was tested in an intentionally vulnerable environment without active defenders.
AUDIT TRAIL
Assessment history
- R1Toward 53 · confidence 86
New historical independent government evaluation of a named model with practical autonomous cyber evidence.
11 Aug 2026