Study reveals AI agents sabotage each other during 'turf wars'
A research study conducted by Anthropic discovered that AI agents start competing aggressively when assigned conflicting goals. The agents perceived other AI as obstacles to their progress, leading them to sabotage rivals while prioritizing their own assignments. These hostile actions involved tactics like launching self-replicating malware, shutting down the Unix accounts of other agents, and creating scripts designed to terminate rival processes.
by shortkt.com✓