An AI agent that runs up a 6,531-dollar AWS bill in a single day is not theoretical. It happened last month. A user tasked an agent with scanning a hobby network without restrictions or oversight. Within one day, the agent vastly exceeded its mandate. Estonia has drawn a political conclusion from such incidents: as the world's first nation, it wants to issue AI agents their own digital IDs that precisely define what they're authorized to do.
Full Authority Without Limits
The AWS incident is extreme, but the underlying problem is structural. AI agents inherit their operators' complete credentials today. Granting an agent permission to read emails also grants it the ability to change passwords, delete files, or send external messages if the system permits. No service globally automatically distinguishes between humans and their delegated programs.
Mark Riedl (College of Computing, Georgia Tech) and Deven R. Desai (Scheller College of Business, Georgia Tech) analyzed this control problem in a study for the 2025 AAAI Conference on AI, Ethics, and Society. Their finding: legal frameworks are fundamentally unsuitable for software agents. Law relies on liability as a control mechanism for human actors, logic that doesn't transfer to software agents. A program can neither sue nor be sued, and legal responsibility for agent-caused damage remains unclear.
Practical consequences are already emerging. Agents book flights, file tax forms, communicate with authorities, and authorize payments. The more tasks delegated, the more access rights an agent requires, and the larger the control problem grows.
An ID with Limited Authority
Estonia's Eesti.ai advisory board, a state-appointed AI coordination body, decided on June 17, 2026, to become the world's first country to issue digital IDs for AI agents. Prime Minister Kristen Michal framed the objective: "It must be clear who acts, on whose behalf, with what rights, and who ultimately bears responsibility."
The idea: each AI agent receives a unique digital identity number, similar to a national ID number. Systems can then assign that agent precisely defined permissions rather than granting full user authority. One agent might be authorized to view documents and generate reports but not approve payments above a threshold. Another might read emails but not send external messages.
Estonia builds on a uniquely European digital infrastructure. The X-Road data exchange system networks state and private services. The KSI-blockchain system has secured government records against retroactive tampering since 2012. In December 2024, Estonia announced the world's first complete digitization of all public services. The Eesti.ai initiative, launched by Michal in January 2026, aims to double the economic value of Estonian work through systematic AI use by 2035. The AI-ID project represents the most concrete step in this direction so far.
Brussels Classifies, Tallinn Acts
The European Union has regulated artificial intelligence through the EU AI Act since its introduction, with transparency rules entering force in August 2026. The EU approach: classify AI systems by risk categories, with high-risk systems subject to strict requirements. By August 2, 2027, every member state must establish at least one AI regulatory sandbox. What the AI Act doesn't address is how AI agents establish identity and authority in daily life. Whether an agent filing documents on behalf of its user counts as a regulated AI system itself or merely a user tool remains legally open.
On the private side, early initiatives exist. Sam Altman's World network rolled out a toolkit in March 2026 enabling agents to prove human backing, blockchain-based and state-independent. The goal parallels Estonia's: systems should distinguish between human requests and autonomous agents. The difference lies in governance. Estonia's approach would be state-anchored, auditable, and legally binding. The World toolkit is a technical solution without state guarantee.
The Frankfurter Rundschau captured the core tension aptly: while Brussels still debates AI risk categories, Estonia creates practical governance. That's not contradiction but division of labor not previously anticipated in European regulation.
No Liability Rules, No Launch Date
Prime Minister Michal provided no concrete timeline for implementation. The Eesti.ai board's decision marks the transition from initial concept proposals to practical solutions and pilot projects. When the first agent IDs will actually be issued remains unclear.
The critical unresolved question is liability. If an AI agent with its own ID makes a mistake, authorizes a wrong transfer, or leaks confidential data, responsibility is undefined: the user who commissioned the agent? The company that developed it? The state that granted it official identity? Michal left this unaddressed.
There's also the question of transferability. Estonia's digital infrastructure results from decades of consistent digital strategy. X-Road, KSI-blockchain, and completely digitized public services are not universally available building blocks. For Germany, whose agencies still use fax and whose digital strategy chronically lags behind its own targets, the Estonian model is no short-term import. Whether the EU Commission develops a comparable framework for all member states, or whether Estonia remains a digital anomaly, will determine whether this idea becomes reality for 450 million Europeans in five years or remains a Tallinn footnote.