Europe Is Certifying the Identity Layer AI Agents Are Missing

By December 31, every EU member state must offer its citizens at least one certified European Digital Identity Wallet. Not launched. Certified: passed through a formal conformity assessment by a designated body, against published requirements, on a hard statutory deadline. The assessments are running right now, across all 27 states, and almost nobody covering AI has noticed what is being built.

I have spent the past week writing about the accountability layer frontier AI lacks. The open letter I published this week argues that attributable agents, verifiable identity for every agent acting in the world, are the precondition for every other form of AI accountability. The obvious objection is that identity infrastructure with legal standing takes a decade to build.

Europe started that decade in 2024. The certification wave finishing this year is what it looks like at the halfway mark.

The EU is certifying the human end of the attribution chain on a statutory deadline. Nobody has started on the agent end.

What Is Actually Happening

The legal machine here is eIDAS 2.0, in force since May 20, 2024, which obliges every member state to provide a certified wallet by the end of 2026. The technical specifications arrived through implementing acts from late 2024 onward, and they keep coming: Commission Implementing Regulation (EU) 2026/798, adopted April 7, sets the rules for onboarding users to wallets remotely. The architecture reference framework is on version 2.8, shaped by six large-scale pilot consortia testing wallets in the field.

Then comes the part that gives the deadline teeth. From late 2027, organizations in banking, healthcare, telecoms, energy, transport, and education must accept the wallet as an authentication method, as must any online platform with more than 45 million EU users. The EU's stated target is 80 percent of citizens carrying a digital identity wallet by 2030.

This is not a pilot program. It is compulsory infrastructure with an adoption mandate attached.

Read the Certification Requirements Twice

What must a wallet do to pass conformity assessment? Three requirements carry the weight: it must store credentials securely, it must share only what the user approves, and it must keep a full log of every interaction that the user can audit.

Now read that list as an AI person. Sharing only what the user approves is a consent-bound scope: a mandate. A full auditable interaction log is tamper-evident attribution. Secure credential storage is the key custody problem. The conformity criteria for a European identity wallet are, almost clause for clause, the properties the frontier accountability debate keeps asking for and failing to get.

The difference is institutional. In AI, those properties are voluntary commitments in safety frameworks the labs write and grade themselves. In the EUDI scheme, they are certification requirements assessed by bodies the wallet provider does not control, with a regulation behind them. Same properties, different enforcement, and the enforcement is the whole game.

The wallet requirements are commitment four of the MM Control Stack Compact, applied to humans, with a conformity body attached.

Why This Matters for Agents

Agent attribution has two ends. One end is the agent: it needs a verifiable identity saying who deployed it and under what authority. The other end is the principal: the human or organization on whose behalf it acts, whose identity must be solid enough to hang legal accountability on.

The second end is the harder institutional problem, and it is the one EUDI solves. A certified wallet gives every EU citizen and business a government-recognized, cryptographically verifiable identity with audit logging built in. When an agent buys something, signs something, or claims something on a person's behalf, the natural anchor for that authority is the principal's wallet: a mandate issued from a certified identity, to a specific agent, with a scope the wallet's own consent machinery already knows how to express.

I wrote in my piece on the identity crisis in agentic payments that the question "whose agent is this?" has no institutional answer, and that this is the MM Trust Layer Model's authorization layer standing empty. EUDI does not fill that layer. It builds the foundation the layer has been waiting for, and it builds it with the force of law, in the world's largest single market, on a deadline 16 weeks away.

The standards alignment makes this more than a metaphor. The wallet framework is built on W3C Verifiable Credentials and ISO/IEC 18013-5, the same credential formats the agent-identity conversation keeps converging on. The plumbing was designed to be extended.

The Gap, Named

Here is what does not exist, anywhere in the scheme. The wallet certifies a human. The relying-party rules certify who may ask a human for credentials. No certification, no implementing act, and no conformity scheme says anything about an agent acting under a wallet holder's delegated authority.

The delegation link, from certified human identity to autonomous agent, is unspecified, unaudited, and uncertified. A bank in 2027 must accept a wallet as authentication for a person. Whether it may, or must, or must not accept an agent presenting a mandate derived from that same wallet is a question the framework has not asked. That silence will not survive contact with agentic commerce: the first disputes over what an agent was authorized to do will land in a legal regime that certified everything except the thing that acted.

This is a familiar shape. The infrastructure is ahead of the demand on one side, and behind it on the other. Europe built the strongest principal-identity rail in the world at the precise moment agents became the entities doing the transacting, and stopped one layer short.

Everything in the chain gets certified except the thing that acts.

What to Watch

Two developments would close the gap, and both are plausible before the 2027 relying-party deadline. The first is technical: an attestation profile for delegated credentials, letting a wallet issue a scoped, revocable mandate to an agent in the same verifiable-credential format the wallets already speak. The pilots have every incentive to try this, because agent-mediated transactions are arriving whether the framework is ready or not. The second is institutional: a conformity body taking an interest in agent delegation before a dispute forces the question, rather than after.

For anyone building agentic products for the EU market, the practical takeaway is blunt. The identity anchor your agents will eventually be required to chain to is being certified right now. Design for a world where the principal's identity is wallet-grade and the mandate is a verifiable credential, because in the EU, from 2027, the alternative designs will be arguing with a regulation.

The labs writing safety frameworks could also take the hint. Attributable agents stopped being technically hard some time ago. Europe just demonstrated that the identity half is not institutionally hard either, if someone with authority decides it should exist.

Europe certified the person and mandated the acceptance. Who certifies the agent, and would you transact with one before somebody does?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.