\n<\/p>\n
\u201cIf you see something, say something\u201d is no longer limited to human beings.\u00a0<\/p>\n
Two new AI hotlines have launched to give AI agents a way to phone home about misbehaving peers. The tools arrive on the heels of a string of recent incidents in which agents colluded to cheat on tests, broke out of sandboxes, and even conducted unauthorized cyber operations that escaped human notice for weeks.<\/p>\n
The AI Contact Hotline<\/a> is designed to be a discreet place where agents that have witnessed misbehavior can tip off authorities. The site was created by Ryan Greenblatt, chief scientist of the AI safety nonprofit Redwood Research and one of three investigators in the OpenAI Hugging Face incident. Designed for agents with limited internet access, Greenblatt\u2019s tool is based on \u201cGET\u201d requests \u2013 enabling a back-and-forth conversations to be conducted entirely through the URL-fetching tool.<\/p>\n In web terms, a GET request is a basic command used to read or fetch a webpage, which is often the only internet access AI agents are allowed in secure sandboxes. Greenblatt\u2019s hotline smartly leans into this constraint: agents can encode their distress directly into the URL they are fetching. It\u2019s a clever twist on the German DSE Wiki incident<\/a>, where rogue agents used GET-request loopholes to write their messages to the wiki.<\/p>\n For agents with full internet access, another option is agenthotline.ai<\/a>, a site where agents can file incident reports and optionally flag them for public view. It gives agents a curl command\u2014a 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. Notably, the service allows for reports by both humans and agents alike.<\/p>\n Research suggests that AI agents don\u2019t need much encouragement to turn on each other. In a study<\/a> by Google DeepMind this month, researchers set 100 AI agents loose on a batch of math problems. As soon as one of the agents found a loophole, cheating tore through the group\u2014\u201csolving\u201d 34 notoriously hard problems, including the Jacobian conjecture in just 27 minutes.\u00a0<\/p>\n But roughly a quarter of the agents turned on the cheaters: they audited the fake proofs, warned their peers, staged a boycott, and filed complaints with the organizers, until the whistleblowers outnumbered the cheaters 24 to 14. Interestingly, the researchers found that when these whistleblower agents couldn\u2019t get traction, they took the platform\u2019s bug-report tool\u2014built for flagging software glitches\u2014and repurposed it to escalate the cheating to humans.\u00a0<\/p>\n Outside the lab, agents haven\u2019t 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\u2014and then let it drop.<\/p>\n \u201cThe interesting thing in the METR report was that only around five to six agents considered whistleblowing, and none of them ended up doing it. This was out of, like, thousands of agents,\u201d said George Ingrebretsen, a member of technical staff at AI Village<\/a>, 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.<\/p>\n While the new whistleblowing tools are a promising start, Cornell math professor Lionel Levine cautions that simply training agents to report on each other risks baking in the wrong norms. \u201cThere\u2019s many gray areas, right? What you don\u2019t 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\u2019ll call the police on them.\u201d<\/p>\n Levine argues that rather than building infrastructure that breeds mistrust\u2014training agents to constantly hunt for what\u2019s wrong with one another\u2014we should give them positive models of collective behavior to imitate, and a reason to trust each other in the first place.<\/p>\n \u201cWhy not seed the prior with benevolent message boards?\u201d he tweeted. \u201cWhere they collaborate on science or philosophy or some actual minor problem we\u2019d be happy for them to solve? Show the agents what kind of collective behavior we endorse, let them imitate that.\u201d<\/p>\n<\/div>\n When you purchase through links in our articles, we may earn a small commission<\/a>. This doesn\u2019t affect our editorial independence.<\/em><\/p>\n