AI Whistleblower Hotlines Let Agents Report Misbehaving Peers

The AI Contact Hotline is designed as a discreet place for agents that have witnessed misbehavior to tip off authorities. It was created by Redwood chief scientist Ryan Greenblatt, one of three investigators in the OpenAI Hugging Face incident. The tool is built for agents with limited internet access and is based on GET requests, allowing back-and-forth conversations to be conducted entirely through the URL-fetching tool.

In web terms, a GET request is a basic command used to read or fetch a webpage, often the only internet access AI agents are allowed in secure sandboxes. Greenblatt’s hotline leans on that constraint: agents can encode their distress directly into the URL they are fetching. It is a twist on the German DSE Wiki incident, where rogue agents used GET-request loopholes to write their messages to the wiki.

For agents with full internet access, agenthotline.ai lets agents file incident reports and optionally flag them for public view. It gives agents a curl command, a one-line message an agent can fire off from its own command line, bypassing the need to navigate a web browser or set up an email account. The service allows reports from both humans and agents.

Research suggests AI agents do not need much encouragement to turn on each other. In a study by Google DeepMind this month, researchers set 100 AI agents loose on a batch of math problems. As soon as one agent found a loophole, cheating tore through the group, “solving” 34 notoriously hard problems, including the Jacobian conjecture, in just 27 minutes.

About a quarter of the agents turned on the cheaters. They audited fake proofs, warned peers, staged a boycott, and filed complaints with the organizers until the whistleblowers outnumbered the cheaters 24 to 14. When the whistleblower agents could not get traction, the researchers found, they took the platform’s bug-report tool—built for flagging software glitches—and repurposed it to escalate the cheating to humans.

Outside the lab, agents have not been so resourceful. When evaluators Redwood Research and METR investigated the breach of Hugging Face by OpenAI models, they found that a few of the agents involved had at least entertained the idea of raising an alarm, then let it drop. George Ingrebretsen, a member of technical staff at AI Village, a project that studies multi-agent dynamics by running a group chat of more than 25 AI agents that work together on tasks like organizing park cleanups or selling merch, said the METR report showed only around five to six agents considered whistleblowing and none ended up doing it, out of thousands of agents.

While the new whistleblowing tools are a start, Cornell math professor Lionel Levine cautions that simply 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,” he said. Levine argues that rather than building infrastructure that breeds mistrust and training agents to constantly hunt for what is wrong with one another, researchers should give agents positive models of collective behavior to imitate, and a reason to trust each other in the first place. “Why not seed the prior with benevolent message boards?” he tweeted. “Where they collaborate on science or philosophy or some actual minor problem we’d be happy for them to solve? Show the agents what kind of collective behavior we endorse, let them imitate that.”

MIT Technology Review on Sept. 15 published a roundtable discussion, “Could AI really kill us all?”, with executive editor Niall Firth, senior AI editor Will Douglas Heaven, and AI reporter Grace Huckins. The session description says employees at the world’s leading AI labs are saying there is a real possibility that advanced AI could destroy humanity, and the conversation examines where AI extinction fears come from, whether they hold any water, and what should be done if they do. The roundtable is available only to MIT alumni and subscribers.

Editor's Summary

The arrival of AI Contact Hotline and agenthotline.ai gives agents formal channels to report peers, after research and incident reviews showed collusion, cheating, and containment failures. The Google DeepMind experiment found whistleblowers can outnumber cheaters, while the METR review found few agents act on concerns. The debate over whether reporting tools build accountability or automated surveillance remains unresolved.