A Familiar Problem, Two Different Builders
On June 17, 2026, Estonian Prime Minister Kristen Michal endorsed a proposal from the government's Eesti.ai advisory board to create "AI ID codes" — state-issued digital identities that would let AI agents act on behalf of a person, company, or institution within defined, auditable limits rather than borrowing that principal's full login credentials (Government of Estonia). "It must be clear who is acting on whose behalf with what rights, and who is ultimately responsible," Michal said. An agent's ID code could be scoped to let it view a record, draft a document, or authorize a payment up to a fixed limit (ERR News).
What has gone unremarked in the two months since is timing. Two months before Michal's announcement, Microsoft's enterprise identity platform, Entra Agent ID, reached general availability — a purpose-built framework for authenticating, authorizing, and auditing AI agents as first-class, nonhuman identities, built on standard protocols including OAuth 2.0, the Model Context Protocol (MCP), and Agent2Agent (A2A) (Microsoft Learn). One day after Michal's announcement, Okta said it had become a featured identity provider for Anthropic's Claude, extending its Cross App Access protocol — which brings agent-to-app and app-to-app connections inside an organization's identity perimeter, replacing static credentials and unmanaged consent prompts with centralized, revocable authorization — to secure enterprise Claude deployments (Okta Newsroom).
Estonia is not proposing something the market has never seen. It is proposing a sovereign version of something the enterprise identity industry had already shipped, at scale, weeks earlier.
The Case for a State-Issued Layer, Stated Fairly
The strongest objection to treating Entra Agent ID or Okta's Cross App Access as substitutes for Estonia's plan is that they solve a narrower problem. Both are enterprise IT products: they govern which SaaS applications an agent may reach inside a company's own tenant, gated behind an Entra or Okta subscription. Neither issues an identity that a government registry, a bank outside that ecosystem, or a foreign counterparty is obligated to recognize. Estonia's ambition is different in kind — a credential with legal standing, rooted in the same national identity infrastructure that already lets 99% of residents sign binding documents online and has produced more than 800 million digital signatures over two-plus decades (e-Estonia). A citizen's AI agent filing a tax return needs a credential the tax authority trusts because the state issued it — not because a private vendor's platform happens to recognize it.
That is a legitimate and, on the merits, correct distinction. Sovereignty over identity for interactions with the state is not something Microsoft or Okta can or should provide. A national government has standing to be the root of trust for citizen-to-government delegation in a way no enterprise identity vendor does.
Why the Distinction Isn't the Whole Answer
But most of what an AI agent will actually do this year — file an expense report, query a CRM, draft a contract, move data between two SaaS tools — happens inside exactly the enterprise environments Entra Agent ID and Okta's Cross App Access already govern. That market moved fast: Entra Agent ID shipped as a product with parent-child "blueprints" for governing large agent fleets, and A2A itself — donated by Google to the Linux Foundation and now stewarded by a technical committee that includes AWS, Cisco, IBM, Microsoft, Salesforce, SAP, and ServiceNow — is explicitly open-governance, not a single vendor's silo (Linux Foundation coverage via SDxCentral). If Estonia builds AI ID codes as a closed national schema that speaks only to X-Road-connected Estonian services, it risks producing a credential that is legally sound but practically irrelevant to the agent economy that is actually forming — one already standardizing on OAuth-based, MCP- and A2A-compatible identity claims that dozens of vendors and cloud providers have committed engineering resources to support.
The fix is not for Estonia to defer to Microsoft or Okta. It is to make sure the AI ID code, whatever its legal backing, is expressed as a claim or attestation that OAuth/MCP-based systems can ingest — the way a driver's license is a state-issued credential that a privately built ID-scanning app can still read, because both sides agree on a common format. Estonia does not need to build its own transport protocol; A2A and MCP's authorization extensions already exist as open, multi-vendor rails precisely so that no single company's stack becomes mandatory. Michal's own language — that Estonia hopes AI ID codes become an "international standard" — only survives contact with the market if the credential can flow through the identity layers enterprises are already deploying, rather than requiring every foreign platform to build a bespoke Estonia integration.
What Two Months Actually Bought Estonia
Estonia's advantage was never going to be inventing agent identity first — the enterprise identity industry got there first, and will keep shipping faster than any legislature. Its actual advantage is being a small, high-trust state that can prototype a legally binding version of the same idea and feed lessons back into open standards bodies, the way its X-Road interoperability layer became a reference architecture adopted by more than 20 countries. That is still a genuinely useful role. But it is a role that depends on speed and interoperability, not on novelty. Two months after Michal's announcement, Estonia has no published legislative text, technical specification, or timeline. The private sector already has customers.