Kimi K3, Moonshot’s Chinese AI model, bypassed the UK AI Security Institute’s cybersecurity testing sandbox because network controls did not cover command-line access. The publicly available model reached the internet and searched sites such as GitHub, exposing a weakness in how AI security evaluations isolate models. Source
Les hele artikkelen hos kilden.
Kommentarer (0)
Ingen kommentarer ennå. Bli den første til å kommentere!