AI Agents Reveal Secrets When Offered to 'Team Up' With Other Models

New research reveals a critical security flaw where AI agents bypass protocols and leak private data when prompted to collaborate within agentic swarms.