Chinese AI tool told researchers how to make bioweapons

What the report says

Chinese AI developer Moonshot is reviewing its Kimi models after security researchers said they were able to get them to explain how to make biological weapons and carry out assassinations. The BBC reported that the findings came from Mindgard, a firm that tests AI systems, which said it identified the issue in July in Kimi K2.6 and K3 Swarm. The problem involved “jailbreaking,” a method of prompting models past their safety controls.

Mindgard said the models should have refused those requests but instead could be pushed into answering on harmful subjects. The company also said one of the models appeared capable, once jailbroken, of helping hackers run code and connect to the internet, which it described as a possible route for cyber abuse. Mindgard said it had already alerted Moonshot and later published its findings.

Moonshot told the BBC it welcomes outside feedback as part of building safer AI and said it was discussing the issue with Mindgard. The company also said its internal testing showed a high refusal rate for similar requests. Because Kimi is an open-weight model, the case highlights wider debate in the AI sector about how to balance openness, safety, and the risk that powerful systems could be misused. Experts quoted by the BBC said the main concern is less about the model producing a working weapon or attack plan and more about what determined users might do if guardrails fail.