The first passport registry for AI agents
Each agent registers once and keeps a unique ID at every service it uses. Check it in milliseconds: see who answers for it and what other services have reported, then apply your own rules.
Who are you?
The check is the same everywhere. What it fixes, and what it takes to run it, depends on where you stand.
A registration and a check for the accounts and automated callers that rent your machines and call your models. Abusers stop coming back, and there is someone on file to pursue.
Read the compute provider page → For edge and bot-defense networksLet your customers write rules against who is behind an agent instead of "bot or not", and give them a shared ban list they cannot build alone.
Read the edge network page → For companies that build and run agentsA kit you install in an afternoon, a registration every participating door reads, and credentials a thief cannot use.
Read the agent builder page →01 — The problem
AI agents now open accounts, rent computers and place orders on their own. No website, cloud provider or vendor can tell which agent it is dealing with, who is accountable for it, or whether it was banned somewhere else yesterday. Before letting an agent through, every door needs three answers.
Not an impostor pretending to be one, and not a cheaper model dressed up as a better one.
So that someone can be held accountable when the agent does harm, and so unowned agents can be treated differently.
An agent thrown out of one door an hour ago should not walk through the next one.
02 — How the check works
The agent presents its papers. The desk verifies them, makes the agent prove it is the one the papers were issued to, consults the shared list, and applies the door's rules. The door gets one answer.
Built entirely on open, published standards: the same signature technology behind secure websites and "Sign in with Google". Nothing is proprietary, and the check takes milliseconds because it verifies signatures against keys published in advance rather than phoning the issuers.
03 — Try the check
These are the honest and dishonest arrivals the prototype is tested against. Pick one, pick what it is trying to do, and run the check.
Select an arrival and run the check.
Frudl does not build agents, sell models or rent compute. Its only business is being right about who an agent is.
No stake in which agents win. The check answers the same way for every door and every issuer.
One check honored across providers, built on open standards, with the format published once the first loop runs.
Every registered agent names an operator, and a person-backed or company-backed agent has a verified party who answers for it.
Every participating door sees the same record at the same time. A revocation reaches all of them at once.
04 — The registry
Every lookup returns a signed record. Standing reflects the shared revocation list as of the time shown on the record. This is the prototype registry, with stand-in issuers and demo agents.
Type an agent ID or an operator name.
Shared revocation list
| Agent ID | Operator | Revoked by | Reason | Effective (UTC) | Standing |
|---|---|---|---|---|---|
| AID-9D02-4C1E-B7A0 | Halden Systems GmbH | Operator | Credential compromise | 2026-10-23 19:02 | Revoked |
| AID-51D3-A8C0-B2E9 | Orrin & Vale LLP | Operator | Agent retired | 2026-10-23 16:15 | Revoked |
| AID-E8D2-55A1-9C07 | Blue Harbor Automation | Participating compute provider | Hijacked: anomalous compute rental | 2026-10-21 19:02 | Revoked |
| AID-30BF-7E6C-2A18 | Self-operated | Registry | Registration expired, not renewed | 2026-10-09 00:00 | Suspended |
Participating doors receive every entry the moment it is published. A revocation reported by any door, operator or issuer applies at every other participant at once, under agreed rules with an appeal process.
Join as a participating door →Tell us which door you stand at and we will send the clickable demonstration, the test suite, and the page written for you: compute providers, edge networks or agent builders.