US Cyber Tests Show Chinese AI Model Kimi K3 Trails Top Rivals

The US government recently put China's newest artificial intelligence model through a series of hacking tests and found that Moonshot AI's Kimi K3 falls well short of top American systems. The Commerce Department shared the results following a joint evaluation with British experts. The tests arrived just one day after Washington accused Moonshot of building Kimi K3 using stolen technology from US firms.

During testing on ExploitBench, Kimi K3 scored 32 percent, which beat China's GLM 5.2 model at 24 percent but trailed top US models that hit roughly 76 percent. Furthermore, the Chinese model failed to take full control of a target machine in all 41 attempts, while leading American systems succeeded on 20 tests. However, analysts note that the American models were tested with safety filters turned off to show peak performance, while some test methods remain private.

Despite the lower scores, researchers warn that Kimi K3 still presents risks because its weak guardrails failed to block hacking attempts. Moonshot plans to release the model to the public soon, meaning anyone will be able to download and modify it. Traders and tech observers are watching closely to see how the global AI race unfolds as Chinese labs continue closing the performance gap.