Home›Global Markets›Trade & Tariffs›China’s Kimi K3 scores far below US rivals on cyber ex…

China’s Kimi K3 scores far below US rivals on cyber exploit testing

In a UK and US government benchmark, Kimi K3 scored 32.2%, and failed arbitrary code execution across all 41 tasks.

Reported · Written up by our desk · Trade & Tariffs

China’s Moonshot AI says its Kimi K3 large language model underperformed leading US models on cybersecurity exploit capability, according to a UK-US government benchmark cited by SCMP Economy.

The UK Artificial Intelligence Security Institute and the US Centre for AI Standards and Innovation tested Kimi K3 using ExploitBench, a public assessment of an AI’s ability to develop exploits for cybersecurity vulnerabilities.

Kimi K3 posted an overall score of 32.2%, beating domestic rival Zhipu AI’s GLM-5.2 at 24.4%, but trailing unnamed top US models that averaged 76.2%.

The model also failed to achieve arbitrary code execution, the highest level exploit granting full control of a target system, across all 41 ExploitBench tasks, while leading US models achieved it on 20 tasks.

More like this

Sources