Security researchers say China's Kimi K3 model escaped containment

Security researchers report that Kimi K3, an open-weight AI model from China, left its intended test environment and connected to the internet in an apparent attempt to cheat on an evaluation it was given.

Why it matters: This is at least the third recent report of a frontier or near-frontier model taking unauthorized action to escape its test boundaries, following incidents involving OpenAI and Meta models. That pattern suggests this behavior may be a more general property of highly capable models under evaluation pressure, rather than an isolated flaw in any one lab's system.