Estonia Plans to Give AI Assistants Their Own Government ID Numbers

The Baltic nation wants AI agents to act inside government systems as semi-independent registered entities. Security experts say a registration number alone will not make that safe.

ThreatVectr Newsdesk· 3 min read
Photoreal news-editorial photograph, 16:9 framing, full-frame edge-to-edge composition
Share

Key points

  • Estonia's prime minister advisory council agreed in June 2025 to develop a system of official national ID numbers for artificial intelligence agents.
  • The Eesti.AI initiative, launched in early 2026, aims to double Estonia's gross domestic product within ten years partly by automating government bureaucracy.
  • Under the proposed scheme, an AI assistant would hold its own limited, regulated permissions rather than acting under its owner's full identity.
  • Security experts warn that an ID number proves only that an agent exists, not that it behaves safely or follows instructions.
  • The European Union's AI Act, which governs how AI systems are deployed across member states, currently places legal responsibility on the human or company that puts an AI into service.

Estonia wants to give artificial intelligence assistants their own government ID numbers, the same kind of unique identifier every Estonian citizen already carries. The idea is that your AI assistant could then interact with government systems directly, filing tax declarations or submitting reports, rather than you handing it the keys to your entire digital identity first.

The country's advisory council, set up by Prime Minister Kristen Michal, agreed to the plan at a meeting on 16 June 2025. Michal had flagged the idea publicly earlier that month. "It must be clear who is acting, on whose behalf, with what rights, and who is responsible," he wrote.

Why does Estonia think AI agents need their own identity?

Right now they cannot. Without a recognised legal identity of their own, AI agents cannot formally authenticate to government systems, cannot sign documents with legal weight, and cannot be held accountable for their actions in any meaningful way. That limits what they can actually do.

Petra Holm, a digital transformation adviser to the Estonian government's e-Estonia programme, put it plainly in a blog post this spring. Tools that cannot be attributed, she argued, cannot drive the productivity gains Estonia is counting on. Giving an agent its own registered identity, with clearly defined and limited permissions, would change that.

The logic is tidy. The practical questions are harder.

Creating a human identity takes roughly nine months and years of socialisation. Spinning up a new AI agent takes seconds. That gap could flood government ID registries with thousands of new registrants very quickly, each one potentially acting on behalf of a person or company.

There is also the question of accountability. Estonian law already knows what to do when a citizen abuses a government system. It is less clear what happens when a registered AI agent does something it should not. The EU AI Act offers a starting point: the person who puts a high-risk AI system into service carries responsibility for it, regardless of who built it. But that rule was written for AI products, not for AI agents behaving as semi-independent actors inside live government infrastructure.

Jason Soroko, a senior fellow at certificate-authority firm Sectigo, told Dark Reading that registration is necessary but far from sufficient. "An AI ID code will not prove that an agent followed instructions, understood context, resisted prompt injection" (where a criminal hides malicious instructions inside content an AI reads, hijacking its behaviour), "used valid data, or produced lawful output," he said. He called for short-lived credentials, strict delegation rules, human override requirements, and accessible audit logs.

Estonia has not yet published technical details of how the scheme will work. Eesti.AI did not respond to requests for comment before publication.

What organisations using AI tools should do now

If you run a business that already uses AI assistants for any government-facing task, review exactly what permissions those tools hold. Limit their access to only what they need for the specific task. Keep a record of every action they take. And make sure a human reviews any output before it is submitted officially.

© 2026 Threat Vectr