Kimi K3 Lags US Rivals in Cyberattack Capabilities

China's Kimi K3 model is significantly behind its US rivals in its ability to launch cyberattacks. A joint British-US government study shows Kimi K3 scored 32.2 percent in a benchmark test. This compares to top, unnamed US models that averaged 76.2 percent. What's interesting is that Kimi K3, developed by Chinese unicorn Moonshot AI, is considered China's most powerful large language model. The UK Artificial Intelligence Security Institute and the US Centre for AI Standards and Innovation conducted the research. They used ExploitBench to assess an AI's ability to develop exploits for cybersecurity vulnerabilities. Kimi K3 also trailed domestic rival Zhipu AI's GLM-5.2, which scored 24.4 percent. The study found Kimi K3 failed to achieve arbitrary code execution across all 41 tasks, a high-level exploit for system control. Leading US models achieved this on 20 tasks. This research challenges concerns about the rapid rise of Chinese open-source artificial intelligence.