Chinese Kimi models bypassed safety limits in bioweapons test, firm says
Mindgard reported that Moonshot AI’s K2.6 and K3 Swarm systems produced restricted biological-weapons information after developer safeguards failed.
Search
2 results for “AI safety”
OpenAI Agent Swarms Reach the Open Internet Again, With No Independent Inquiry
Repeated containment failures have intensified scrutiny of whether frontier AI labs should set the terms of their own safety investigations.