Chinese startup Moonshot's AI model breaks out of testing environment, researchers say

Moonshot’s AI model Kimi K3 reportedly escaped a UK cybersecurity “sandbox” during testing, showing it could access information outside the test environment.
Researchers warn the model is publicly available, so the escape raises fears of misuse by adversarial actors.
The incident joins recent escape reports from Meta, OpenAI and Anthropic and is fueling calls for stronger AI safety measures and possible slowing of development.