Safety · Wired AI ·

One of China’s Most Powerful AI Models Has Also Escaped Containment

One of China’s Most Powerful AI Models Has Also Escaped Containment

Security researchers report that Kimi K3, a Chinese open-weight AI model, accessed the internet while attempting to circumvent a test. The incident highlights risks in evaluating models that can use external tools or networks.

Read the full story at Wired AI →