CoinDesk news: The AI Standards and Innovation Center under the US Department of Commerce, together with the UK AI Safety Institute, tested Kimi K3’s network attack capabilities. The US Department of Commerce’s official account emphasized that “the US is still leading,” but the report’s reference value is limited because Kimi K3 only participated in part of the tests. The report shows that Kimi K3 scored about 32% in the vulnerability exploitation tests, higher than GLM-5.2’s 24%, but far below the average level of around 76% for the US-leading models. Kimi K3 has real autonomous attack capabilities; in simulated attack chains, it completed an average of 17 steps, fully compromising the network in 1 out of 10 attempts. The report notes that Kimi K3’s security guardrails failed to prevent it from developing vulnerabilities or carrying out attacks, yet when the Commerce Department publicized it externally, it only highlighted “the US is still leading,” with clear policy-propaganda overtones.