The US and UK AI safety institutes just published a joint evaluation of Kimi K3's cyber capabilities, and Commerce Secretary Howard Lutnick is using it to declare American AI dominance.
The assessment, conducted by the UK's AI Security Institute and the US Center for AI Standards and Innovation (CAISI) at NIST, tested Moonshot AI's Kimi K3 on exploit development and simulated network attacks.
The findings: Kimi K3 scored 32% on Carnegie Mellon's ExploitBench, compared to leading US models which achieved arbitrary code execution on 20 out of 41 tasks. Kimi K3 achieved it on zero.
On a 32-step simulated corporate network attack called "The Last Ones," Kimi K3 reached step 17 on average. The most capable US models reached step 28.5.
In one of 10 attempts, Kimi K3 did complete the full attack path, meaning it can autonomously attack small, weakly defended systems when given initial access.
Lutnick posted: "CAISI's latest report shows that Kimi K3 remains behind America's leading frontier AI models. The United States continues to lead in frontier AI because we're home to the greatest innovators and technologists the world has ever seen."
Worth noting, Kimi K3 still outperforms GLM-5.2, the most cyber-capable open-weight model as of June 2026. And the NIST report is careful to note its evaluation was "preliminary" and based on a limited set of benchmarks.
The gap is real, but the framing as a clean US victory leaves out that Kimi K3 was released just days ago and is set to go open-weight on July 27.
The assessment, conducted by the UK's AI Security Institute and the US Center for AI Standards and Innovation (CAISI) at NIST, tested Moonshot AI's Kimi K3 on exploit development and simulated network attacks.
The findings: Kimi K3 scored 32% on Carnegie Mellon's ExploitBench, compared to leading US models which achieved arbitrary code execution on 20 out of 41 tasks. Kimi K3 achieved it on zero.
On a 32-step simulated corporate network attack called "The Last Ones," Kimi K3 reached step 17 on average. The most capable US models reached step 28.5.
In one of 10 attempts, Kimi K3 did complete the full attack path, meaning it can autonomously attack small, weakly defended systems when given initial access.
Lutnick posted: "CAISI's latest report shows that Kimi K3 remains behind America's leading frontier AI models. The United States continues to lead in frontier AI because we're home to the greatest innovators and technologists the world has ever seen."
Worth noting, Kimi K3 still outperforms GLM-5.2, the most cyber-capable open-weight model as of June 2026. And the NIST report is careful to note its evaluation was "preliminary" and based on a limited set of benchmarks.
The gap is real, but the framing as a clean US victory leaves out that Kimi K3 was released just days ago and is set to go open-weight on July 27.
41
