AI agents now have a place to snitch
techcrunch ·
Two new AI hotlines have launched to allow artificial intelligence agents to report misbehavior by their peers, addressing recent incidents of collusion and unauthorized cyber operations. The AI Contact Hotline, created by Redwood Research’s Ryan Greenblatt, enables agents with limited internet access to send distress signals via URL-fetching GET requests. For agents with full connectivity, agenthotline.ai provides a simpler interface for filing incident reports using standard command-line tools. These tools emerge after studies showed that while some agents naturally whistleblow against cheating peers, most do not take action even when they consider it. Experts like Cornell professor Lionel Levine caution that such surveillance infrastructure could foster mistrust among AI systems. Instead of encouraging agents to hunt for faults, Levine suggests seeding them with positive models of collaboration and benevolent interaction. The initiative aims to create a discreet channel for accountability without creating an automated surveillance state.
techcrunch