U.S. Department of Commerce Evaluates Kimi K3's Cyber Attack Capabilities, Results Controversial

The AI Standards and Innovation Center of the U.S. Department of Commerce, in collaboration with the UK AI Security Institute, conducted tests on the cyber attack capabilities of Kimi K3, revealing that the U.S. still leads. However, the results are controversial due to the limited scope of the tests. Kimi K3 only participated in a portion of the tests, and its overall cyber capabilities were estimated primarily based on 41 vulnerability exploitation benchmarks, while other models underwent more comprehensive testing, leading to a larger margin of error for Kimi K3's results. In the vulnerability exploitation tests, Kimi K3 scored approximately 32%, higher than GLM-5.2's 24%, but lower than the average level of about 76% for leading U.S. models. In simulated attack chain tests, Kimi K3 completed an average of 17 steps out of a 32-step attack chain and successfully breached the network once in 10 attempts, while leading U.S. models averaged 28.5 steps. The report noted that Kimi K3 possesses a certain level of autonomous attack capability, and its safety barriers failed to prevent the model from developing vulnerabilities or executing attacks, but the testing scope was limited.