[UK Artificial Intelligence Safety Institute Assessment Shows Kimi K3 Cyberattack Testing Lags Behind U.S. Models] According to a joint evaluation by the UK Artificial Intelligence Safety Institute and the U.S. Artificial Intelligence Standards and Innovation Center, Moonshot AI's Kimi K3 achieved a success rate of 32.2% in the ExploitBench vulnerability exploitation benchmark test, higher than GLM-5.2 but lower than the leading U.S. model's 76.2%. Kimi K3 failed to reach the arbitrary code execution level during testing and completed the full process only once out of 10 attempts in simulated enterprise intranet attack tests.