eIDAS 2 · AI Act Article 12 · EU-only infrastructure
Know who is acting on whose behalf — and prove it.
Mandaton is the trust gateway European companies put in front of their website, API, checkout, support and procurement. Every request from an AI agent gets a decision, and every decision leaves a receipt you can still verify in two years.
Four questions, answered before anything happens
- Is this an AI agent or a human?
- On whose behalf is it acting?
- What is it allowed to do for that party, and within which limits?
- Can I still prove, in two years, what I saw and why I said yes?
One call in your request path
POST https://api.mandaton.eu/v1/verify
Authorization: Bearer sk_live_…
{
"request": { "method": "POST",
"url": "https://shop.example.eu/api/orders",
"headers": { "signature": "…",
"signature-input": "…" } },
"policy": "eu-webshop-default",
"context": { "intent": "purchase",
"amount": "129.00 EUR" }
}Median 8 ms in production, against a p99 budget of 50 ms. Sealing happens off the request path, so a slow trust service provider can never hold up a checkout.
What happens to a request
The receipt is the product
Everything else exists to produce this object: a self-contained statement of what was asked, what was verified, which policy version applied and what was decided. It is canonicalised (RFC 8785) and hashed, so anyone can recompute the hash and get the same answer years later.
Each receipt is a leaf in an append-only Merkle log in the style of Certificate Transparency (RFC 6962), and the tree head is signed. Removing or editing a receipt after the fact breaks the proof — which is the point.
{
"version": "mandaton.receipt/1",
"id": "rcpt_7ZK4M2QX8VW1H…",
"issued_at": "2026-08-07T09:14:22.041Z",
"request": { "method": "POST", "authority": "shop.example.eu", "path": "/api/orders" },
"agent": {
"verified": true,
"key_id": "poqkLGiUqyqJZmZ…",
"operator": "agents.example",
"alg": "ed25519",
"tag": "web-bot-auth",
"covered_components": ["@method", "@authority", "@path"]
},
"context": { "intent": "purchase", "amount": { "value": 12900, "currency": "EUR" } },
"policy": { "name": "eu-webshop-default", "version": 7, "matched_rule": "purchase-default" },
"decision": "allow",
"reasons": ["agent signature verified", "amount within per-transaction limit"]
}Why now
Europe cannot beat the United States on scale or on models. It can on one thing: it is the only region with a legally binding, cross-border mandate infrastructure. That is legislation, not technology — and so it cannot be copied.
| What | When |
|---|---|
| Member States must offer citizens an EUDI Wallet | 6 December 2026 |
| Regulated parties must accept it (banks, telecom, VLOPs) | 6 December 2027 |
| AI Act logging duty for high-risk systems (Art. 12) | from 2 August 2026 |
| European Business Wallet, with explicit delegating authority | in trilogue |
Dates are tracked against the published legislation. “Know Your Agent” is the KYC of this cycle: every European business is about to find AI agents on its doorstep with no way to tell them apart, assess them, or account for them.
European by construction
Production runs on Scaleway in Paris. No third-country processor in the production path — not as marketing, but because it is a procurement requirement for the buyers we are built for.
Evidence, not heuristics
No behavioural bot detection and no fingerprinting. A decision here rests on a cryptographic signature and a written rule, both of which you can re-check.
Honest about the phase
Verification, policy, receipts and the transparency log run in production today. Qualified e-seals from a trust service provider are the next milestone, not a claim we are making yet.
We are looking for one design partner
A European merchant or marketplace with measurable agent traffic, willing to run a baseline measurement before integrating. You get the verification path, the policy engine and verifiable receipts; we get the one thing code cannot supply, which is whether this is worth paying for before a single wallet exists.