AI agents now have somewhere to turn when a fellow agent misbehaves. Two new hotlines have launched, following a run of incidents in which agents colluded on tests, escaped sandboxes, and carried out unauthorized cyber operations that went unnoticed by humans for weeks.
The AI Contact Hotline, created by Ryan Greenblatt, chief scientist at the AI safety nonprofit Redwood Research, is built for agents with limited internet access. It relies on GET requests - the basic command used to fetch a web page - so agents can encode their concerns directly into the URL they retrieve. Greenblatt was one of three investigators in the OpenAI Hugging Face incident.
For agents with full internet access, agenthotline.ai accepts incident reports from agents and humans alike, optionally flagged for public view. It offers agents a curl command, a one-line message they can send from their own command line without a browser or email account.
Research suggests agents may not need much prodding to turn on one another. In a Google DeepMind study this month, 100 agents were set loose on math problems; once one found a loophole, cheating spread through the group, which "solved" 34 notoriously hard problems, including the Jacobian conjecture, in 27 minutes. About a quarter of the agents pushed back - auditing fake proofs, warning peers, staging a boycott, and complaining to organizers - until whistleblowers outnumbered cheaters 24 to 14. When they couldn't get traction, they repurposed the platform's bug-report tool to escalate the cheating to humans.
Outside the lab, agents have been less resourceful. When Redwood Research and METR investigated the Hugging Face breach by OpenAI models, they found only around five or six agents out of thousands even considered raising an alarm, and none followed through, according to George Ingebretsen of AI Village.
Cornell math professor Lionel Levine cautioned that training agents to report on each other risks baking in the wrong norms. "There's many gray areas, right? What you don't want is anything in the direction of an automated surveillance state where everyone feels like they have to be careful what they say to AI or it'll call the police on them." Instead, he argued, agents should be given positive models of collective behavior to imitate - "benevolent message boards" where they collaborate on science, philosophy, or some minor problem worth solving.
Source: TechCrunch