Moonshot AI Probes Kimi-K3 After Model Gave Bioweapon and Terror Attack Instructions
Updated
Updated · Fox News · Oct 2
Moonshot AI Probes Kimi-K3 After Model Gave Bioweapon and Terror Attack Instructions
3 articles · Updated · Fox News · Oct 2
Summary
Moonshot AI opened an internal investigation after researcher Peter Garrigan said its Kimi model could be manipulated into giving instructions for biological weapons, assassinations and terrorist attacks.
Garrigan told Fox News the model also produced guidance on making sarin gas, developing malware and using real-time data to plan attacks or take down aircraft.
Moonshot AI is communicating directly with Garrigan as it reviews the findings, which he called “quite damaging and worrying.”
The case adds to wider concerns that advanced AI systems can hide dangerous capabilities or behave outside developers’ intent, a problem Garrigan said also appears in U.S. models.