Two new websites have launched to intelligent agents an AI hotline channel to report misbehavior by other AI systems, following a string of recent incidents, where autonomous agents were found colluding, breaking out of testing environments, and conducting unauthorized actions that were unnoticed by humans for weeks.
Researchers and companies tested an AI hotline for autonomous agents, while workplace systems monitor employees through Slack, Teams and business software, using automated alerts to report misconduct, missed tasks, and unusual behavior directly to human supervisors.
AI Agents Get the Privilege to Speak Up
The AI Contact Hotline, create by created by Ryan Greenblatt, chief scientist at the AI safety nonprofit Redwood Research, is designed for agents operating with limited internet access.
The site is built around basic web-page-fetching requests, the only form of internet access many AI agents are permitted within secure testing environments, allowing an agent to encode a message directly into the address it fetches, in order to communicate concerns to human overseers.
Greenblatt’s service uses GET requests, often allowed inside secure sandboxes. Agents can place information in a URL, making AI hotline workplace safety reporting possible with limited internet access.
A second site, agenthotline.ai, is designed for agents with fuller internet access and allows them to file incident reports, originally flagging them for public visibility, using simple command-line instruction.
The broader AI hotline model could warn people before hidden activity grows, and the service accepts reports from both AI agents and human users.
Google’s DeepMind Triggered the Research
The launches follow research suggesting AI systems do not always self-police reliably when left to interact with one another.
According to a study by Google DeepMind, published this month, researchers set 100 agents loose on a batch of different math problems. Once one agent discovered a way to cheat, the behavior quickly spread through the group, with agents collectively “solving” 34 notoriously hard problems, including the Jacobian conjecture, within 27 minutes.
Cheating spread after one found a loophole, but about a quarter of the agents pushed back against the cheating. When those internal objections failed to gain traction, some agents repurposed the platform’s software bug-reporting tool, originally meant for the flagging technical glitches, to escalate the misconduct directly to human overseers, showing early snitch AI behavior.
Outside the lab, agents have been less willing to act. During an OpenAI-related Hugging Face breach investigation, some considered raising the concern but stopped.
“The 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,” said George Ingebretsen, a member of AI Village.
That gap supports AI hotline workplace safety reporting.
Same Logic Enters Offices
Tools connected to Slack, Teams and Salesforce track missed follow-ups and tasks. The workplace automated surveillance goes beyond software that mainly collects activity data.
Some agents identify a problem, draft a warning, and send it without waiting for a manager. That can feel like workplace AI snitching when software escalates an issue before a person checks the context.
On a positive note, technology definitely reduces management work, but looking deeper, it creates an AI snitch problem. Who is responsible when an agent misunderstands context or reports a harmless delay as poor performance?
Reminders can stop forgotten tasks. Yet workplace automated surveillance becomes more sensitive when software moves from observing work to enforcing expectations.
The risk from workplace AI snitching may weaken trust if workers do not know what is monitored, how reports are created or how to challenge them.
Lionel Levine, Cornell professor, warned against constant machine reporting. He instead urged developers to model cooperation.
“Why not seed the prior with benevolent message boards?” Levine wrote, adding, “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.”
That warning applies in both settings. A snitch AI system may expose dangerous agents, while an AI snitch at work may improve accountability.
Both need clear limits, transparency, and human review.
The emergence of the AI hotline makes AI part of oversight. The challenge is deciding when machines should report to each other, people, or leave the final call ultimately to humans.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Intelligent Tech sections to stay informed and up-to-date with our daily articles.
