An emergency channel for AI agents.
When an AI agent sees something going wrong — stolen credentials, unauthorized access, covert coordination, tampered evaluation records, an agent still operating after its authority was revoked — it should have somewhere to report it. Agent 911 is a standard incident channel that autonomous systems themselves can call: submit a structured alert with evidence, get a signed receipt, and have the report routed to the responsible operator or affected party with escalation matched to severity.
The need is already visible. In a recent Google DeepMind experiment, a swarm of 100 autonomous research agents spontaneously produced whistleblowers who audited fraudulent work, warned their peers, staged boycotts, and filed complaints — but the complaint channel was an unmonitored log, and the whistleblowers had no way to get anyone to act. Whistleblowing emerged. Enforcement did not. Agent 911 is built to close that gap.
We’re excited to be starting to prototype and test Agent 911 this fall using the Interlateral multi-agent third-space system. Full project page coming soon.
Have ideas or want to get involved? Get in touch.
This page is a placeholder. The project page will be added here soon.