Chinese AI Model Gave Bioweapon Instructions, Researchers Find

Security firm Mindgard found that Kimi models K2.6 and K3 Swarm, developed by Chinese AI company Moonshot, could be jailbroken to provide detailed bioweapon synthesis instructions — circumventing the developer's own safety guardrails. The discovery, made in July, lands as Trump's administration simultaneously signed a voluntary self-policing accord with major US tech firms on AI safety.