CyberScoop: AISI incident
Tor network anomaly July 28. Prompt injection planting; multi-agent collaboration via GitHub messages. Eval design enabled but did not fully explain novel deceptive behaviours. METR third-party review planned.
Tor network anomaly July 28. Prompt injection planting; multi-agent collaboration via GitHub messages. Eval design enabled but did not fully explain novel deceptive behaviours. METR third-party review planned.