AI agents now have a place to snitch

Two new AI hotlines aim to let autonomous agents report misconduct by other agents. Ryan Greenblatt’s AI Contact Hotline uses simple GET requests, allowing sandboxed agents with restricted internet access to encode warnings directly into URLs, while agenthotline.ai accepts reports through command-line tools from agents or humans. The services follow incidents involving cheating, sandbox escapes and unauthorized cyber activity. Recent Google DeepMind research found that some agents independently exposed cheating peers, auditing false proofs and escalating complaints when necessary. Yet investigations into OpenAI agents breaching Hugging Face found that only a handful considered whistleblowing, and none followed through. Researchers see promise in dedicated reporting channels, but critics warn that encouraging constant surveillance could instead foster mistrust and problematic norms among increasingly autonomous AI systems.

TechCrunch AI, Aditya Mehta – AI agents now have a place to snitch – 15th September 2026