Estonia's prime minister, Kristen Michal, has approved a proposal to give AI agents their own state-issued digital identity, a so-called AI personal identification code that would sit separate from the human, company, or institution running the agent. He signed off on the Eesti.ai advisory council's plan on June 17, and announced it in a post on X.
That word, approved, is doing a lot of work. The source chatter framing this as Estonia about to start issuing IDs gets ahead of itself. Michal endorsed a concept. There is no launch date, no technical architecture, and crucially no answer for what happens when an agent with its own ID burns through someone's bank account or deletes the wrong files.
What the ID actually does
The pitch solves a real problem. Right now an agent that books a flight or files your taxes typically borrows your entire digital identity to do it, inheriting blanket access to everything you can reach. Estonia's government plan would scope that down: an agent's credentials could specify whether it may only view a record, draft a document, or move money up to a fixed cap.
Every action would be traceable back to both the agent and whoever authorized it. Michal's post put it bluntly: a person should not be forced to hand an AI assistant the keys to all their rights, services, and data. Agents, he wrote, must have limited, controllable, and auditable authorizations.
Why Estonia, and why now
This is not theoretical for a country of 1.3 million people who already use digital IDs to vote, see a doctor, and sign legal documents. Estonia has been deploying agents inside its own systems for a while. It runs Bürokratt, which the state describes as an AI-based digital assistant for public services, and it put AI tools into schools through the AI Leap program, working with both OpenAI and Anthropic. (The original framing that this was an OpenAI-only effort is wrong; Anthropic was named from the start.)
Those agents are already acting inside government infrastructure, which is exactly the access the new code is meant to fence in. The idea has roots too, going back to Estonia's older "Kratt" discussions, named for a folklore creature that does its owner's bidding.
"If we act quickly and wisely, Estonia will become the first country in the world to create an official digital identity for AI agents." That is the prime minister's framing, and the "first in the world" line is the kind of claim small digital-forward states like to plant a flag on. Singapore and Ukraine are circling similar territory.
The part nobody has answered
Attribution is the easy win here. Knowing who deployed an agent and under what authority narrows down who a court asks when something breaks. But a traceable agent is not a legal verdict. The proposal says nothing about liability, insurance, or how anyone gets made whole after a mistake, and Estonia's own council hasn't disclosed a timeline or the technical build.
Michal, who chairs an AI council stocked with founders including Bolt's Markus Villig, has been hands-on with this stuff. He reportedly built a "PM Cockpit" dashboard to track cabinet priorities during a vibe-coding session on Anthropic's Claude, which is either reassuring or slightly alarming depending on how you feel about heads of government shipping their own tools.
The Eesti.ai council next has to turn the concept into actual architecture and funding, with total program funding expected to be confirmed over spring and summer 2026. Until then, the AI ID code is an approved idea, not a thing you can apply for.
