Passport ControlAI Passport Ideathon · Egoist Machines
Today, an AI Passport gives agents context and controlled access to things like your repos and calendars, so they can act on your behalf without starting from scratch. That’s useful. But it’s not all a passport is capable of.
The story
I work in the AML department at Utmost, in life insurance fraud detection. Every day I verify passports for accuracy and validity: whether the person in front of me is who they claim to be. Is the date of birth consistent? Does the nationality line up? Even place of birth can shift trust, because certain combinations are exactly what fraud looks like up close.
That same problem is coming for AI agents, and nothing today solves it.
The gap
Agents will take on more: managing transactions, moving data, running parts of businesses. When one agent asks another to act with real consequences, the other side needs to know a real, accountable human sits behind the request.
The same is true when money isn’t involved. Agents shouldn’t hand over data or converse freely in situations that call for knowing who’s on the other side. That’s what passports do for humans. There’s still no agreed way to do this between agents.
For that to work like a real passport, the company behind AI Passport would need to verify people themselves, not just store preferences. Identification has to mean a real human stands behind the agent.
AI Passport already has the infrastructure (Connections, an Inbox, user-owned data). Identification is the missing piece.
How it could work
Agents in conversation or mid-transaction hand over a key or link to their credentials. That credential ties to verifiable information: an agent ID plus a rotating secret required to produce it, so replay and impersonation become effectively impossible. On access, you get a small JSON payload at most: proof that identification was given, plus only the fields the user chose to expose.
Agent A
Hands over a key or link mid-transaction
Agent B
Reads only what the user chose to expose
Not full data. Just proof of claim.
It shouldn’t be rigid. Agents set their own bar. A business agent moving large sums might require a verified name and location. A casual personal agent might need almost nothing. The threshold is set by whoever takes the risk, not a central authority.
With enough adoption, agents that want to be trusted and dealt with will need an AI Passport. That becomes a snowball: each verification ask advertises the passport to another agent, websites join the same standard, and the network gets harder to ignore. Websites could use the same layer too: refuse unverified bots, and let verified, accountable agents through instead of blocking everything with a CAPTCHA.
Try it
Your agent asks a payments service to proceed. Watch identity get challenged, presented, and verified. Open the full scenario demo.
Agent A · Your agent
Ready to request
Agent B · Payments service
Listening for requests
Why this matters now
Bots are about to be everywhere: negotiating, transacting, talking to each other, with no reliable link back to who they belong to or who’s accountable. Without something like this, there’s no trail when one of them gets it wrong.
Passports solved this for humans centuries ago. Agents need the same layer, and nobody owns it yet. For website verification especially, a universal system has to be established so sites can check the same credential standard instead of inventing their own, or falling back to blocking every bot with a CAPTCHA.
Next steps
Open the full demo to walk through payments, agent-to-agent trust, and website verification.
AI Passport Ideathon · Egoist Machines