AI Passport

Inspiration

Every AI assistant we use starts from zero. Tell ChatGPT you hate red-eye flights, tell Claude your writing style, tell a third app your allergies — and the next one has no idea. That's the exact problem Egoist Machines built AI Passport to solve. Asked where that problem gets sharpest fastest, and landed on AI agents that act for you, not just chat with you. Auto-apply and job-screening agents are already live, and they need the most consequential data a person has — salary floor, visa status, references. If a passport can't be trusted to govern that handoff safely, it can't be trusted anywhere agents act on someone's behalf. So I built the passport doing exactly that.

What it does

Agent Border Control is a working mockup of AI Passport governing a job search. A candidate's employment data — skills, salary range, availability, work authorization, references — lives in structured passport fields, never duplicated permanently into any single tool. When an agent (like a job-search assistant) needs data, it requests specific fields, states its purpose, and asks for a time-boxed window. The candidate approves, narrows the scope, or denies. If the agent learns something new mid-task, it proposes a write back to the passport instead of saving it silently — the candidate approves or rejects before anything is stored. Every grant, denial, write, and revoke is logged in real time to a plain-language receipts ledger, and any connection can be revoked instantly.

How we built it

A single self-contained HTML/CSS/JS mockup — no backend, no dependencies — so it opens instantly for judges with nothing to install. State (passport fields, pending requests, proposed writes, active connections, and the receipts ledger) lives in-memory and updates live as you click through the flow. We designed it visually around a "border control" metaphor — ledger pages, stamped approvals, a receipts log — so the interaction itself teaches the concept: nothing crosses without a scope, a duration, and a receipt.

Challenges we ran into

The hardest part wasn't the interface, it was picking the right scenario. Our first instinct was to invent a whole marketplace of specialized agents, which felt inventive but speculative — a judge could reasonably ask "does this problem exist yet?" We rewrote the framing to ground it in what's actually true today: agents doing job search already exist, and the data they need is already the most sensitive category a person has. Narrowing to one believable, present-tense scenario made every other design decision — which fields, which requests, which risks — much easier to justify.

Accomplishments that we're proud of

Every core AI Passport mechanic — scoped read, stated purpose, time-boxed grant, proposed write, revoke, receipt — shows up as something you can actually click, not just describe. The scope-narrowing interaction (trimming a request down before approving it) and the proposed-write approval flow were the two pieces we cared most about getting right, since they're what separate a real permission system from a simple on/off toggle.

What we learned

The trust problem in AI agents isn't really about whether to share data — it's about the absence of a receipt. Once every action leaves a visible, timestamped trail the person can review, the calculus around sharing sensitive data changes completely. We also learned that grounding a "future of AI" idea in something already live today (auto-apply tools exist right now) makes the pitch far more convincing than a more ambitious but hypothetical scenario.

What's next for AI Passport

Extend Agent Border Control beyond job search to any agent acting with financial or legal consequence — apartment applications, loan pre-approval, insurance quotes — anywhere an agent needs high-stakes personal data and a person needs proof of what it did with it. Longer term: a standard "purpose taxonomy" agents declare against (so "apply to jobs" and "negotiate salary" are distinguishable scopes, not one blanket grant), and a real backend so receipts and revocations persist across sessions instead of resetting each demo.

Built With

• agents
• ai-passport
• css3
• egoist-machines
• html5
• javascript
• permissions