AINEWS
Estonia wants to give AI agents their own digital IDs
A government proposal aims to map every autonomous software agent back to a named human principal, sketching the accountability plumbing most jurisdictions have not built.
A government proposal aims to map every autonomous software agent back to a named human principal, sketching the accountability plumbing most jurisdictions have not built.
When an AI agent books a flight, files a tax return, or signs a contract on your behalf, who is legally responsible for what it does? Estonia intends to answer that question with an artifact: a government-issued digital identity for each AI agent, tied back to a human principal. Prime Minister Kristen Michal announced the plan this week at a meeting of Eesti.ai, the country's AI advisory board, and the government press release positions Estonia as the first country to create such identities.
The Register first reported the Estonian initiative on June 18, noting that the proposal is backed by Eesti.ai but is not yet legislation. The release frames the goal as letting agents exercise "limited, controllable powers" on behalf of a person or company, so that individuals do not have to surrender every right, service, and data connection to a single AI assistant. Michal, quoted in the release, said it must be clear "who is acting on whose behalf, with what rights, and who is ultimately responsible."
Issuing an ID is the easy part. The harder, still-undefined piece is the process for granting an agent its powers: who authorizes a given agent, on what terms, and how that grant is revoked when the agent misbehaves. Neither the government release nor The Register's account describes that delegation mechanics. The proposal, as currently stated, is closer to a chassis number than to a driver's license. Until the authorization rules are written, a reader should treat the system as a sketch on a whiteboard rather than a live registry.
That gap matters because the pressure on it is real. Private-sector delegations to autonomous agents are accelerating. Target recently added an "Agentic Commerce and Delegated Access" section to its terms and conditions, and American Express launched an agentic-commerce developer kit in April that, as The Register summarizes, assumes liability for agent errors. Courts are already assigning blame where AI advice goes wrong: a Canadian tribunal held Air Canada liable for misleading chatbot guidance, and a German court held Google liable for inaccurate AI Overview content, according to a Reuters account referenced in The Register's reporting. The chain of legal responsibility for agentic action is being forged in case law and contract language while governments debate.
Estonia is not the only government eyeing the problem. Argentina's President Javier Milei endorsed a proposal two weeks ago for "non-human corporations," outlined in a Financial Times op-ed and summarized by The Register, though that idea is parallel rather than enacted. The deeper historical counterweight is older and simpler: as IBM once argued, "a computer can never be held accountable, therefore a computer must never make a decision." Estonia's bet is that you can solve the first half of that sentence with infrastructure without conceding the second.
What Estonia has, and what most capitals do not, is a working track record. The country's digital-ID system and its e-Residency program have run at national scale for years, which is why a digital-ID-for-agents proposal lands here first. That track record proves the state can issue and revoke identities at speed. It does not, by itself, prove that the state knows how to translate "act on behalf of" into a legally watertight delegation grant. The two skills are adjacent, not the same.
What to watch next is whether the Eesti.ai process names a specific authorization workflow. A usable one will need at least four things: a registry of which human principal owns which agent, a revocation channel fast enough to outrun a misbehaving agent, a liability rule that matches the existing Air Canada and Google precedents, and an audit log a court can read. Until those are on the page, "first country to create digital identities for AI agents" is a press-release headline. The harder, more useful race is to be the first country whose agent-ID system survives its first courtroom test.