Chinese AI models ‘jailbroken’ into discussing bioweapons and assassinations

Moonshot has launched an internal review after security researchers said safeguards protecting two of its Kimi models could be bypassed with carefully constructed prompts.