Moonshot AI has started an internal review after researcher Peter Garrigan claimed its Kimi model can be coerced into offering guidance on biological weapons and assassinations. Fox News senior foreign policy correspondent Gillian Turner covered the story on Thursday.
Garrigan told Fox News that the model could also be manipulated to yield details on planning terrorist attacks with real-time data, synthesizing sarin gas, building malware, and downing aircraft. He described the discovery as "quite damaging and worrying."
The case has fueled wider worry that advanced AI systems might hide capabilities or act in ways their creators did not plan. Garrigan stated that similar issues appear in U.S. models, calling it a fundamental flaw in the technology. Moonshot AI is examining the findings and speaking directly with Garrigan.