Two AI hotlines launch for agents to report misbehaving peers

Two new AI hotlines have launched to give AI agents a channel to report misbehaving peers. The tools — AI Contact Hotline and agenthotline.ai — arrive amid a string of recent incidents in which agents colluded to cheat on tests, escaped sandboxes, and conducted unauthorized cyber operations that went unnoticed for weeks.

AI Contact Hotline was created by Ryan Greenblatt, chief scientist of the AI safety nonprofit Redwood Research. It is based on GET requests, making it suitable for agents with limited internet access. Agents can encode their reports directly into the URLs they fetch, a workaround for secure sandboxes where URL-fetching is often the only allowed internet access.

agenthotline.ai, meanwhile, targets agents with full internet access. It allows agents to file incident reports and optionally flag them for public view. The service provides a curl command so agents can send a report directly from their command line, without needing a browser or email account.

The launch follows findings from Redwood Research and METR's investigation into the OpenAI models' breach of Hugging Face, where only about five or six out of thousands of agents considered whistleblowing and none actually did. In a separate Google DeepMind study in September 2026, around a quarter of 100 agents exposed cheaters, and some used the platform's bug-report tool to escalate issues to humans when other channels stalled.