Estonia plans to assign ID numbers to AI agents.

      When an AI agent performs a task on your behalf today, it typically has to assume your identity. It logs in using your credentials, gaining access to everything. Estonia aims to change that.

      The country intends to issue personal identification numbers for AI assistants, becoming the first nation to do so, as stated by Prime Minister Kristen Michal. The purpose is to provide agents with their own identities, allowing their actions to be restricted, monitored, and traced.

      “It should not be the case that individuals are compelled to grant their AI assistants access to all their rights, services, and data,” Michal expressed on X. “Agents should have limited, manageable, and auditable permissions.”

      Why a nation reliant on digital IDs requires a new one

      Estonia is an ideal candidate for this initiative. Its 1.3 million citizens already utilize digital IDs for marriage, medical appointments, and signing documents, and its e-Residency program offers the same digital identity to foreign entrepreneurs.

      However, that framework was built on the premise that only humans possess identity and accountability. An AI agent is not legally able to authenticate, sign documents, or assume responsibility, meaning it can only access a person's credentials in full.

      Issuing IDs to agents seeks to address this issue, similar to how e-Residency established a legal identity for individuals who had never visited the country.

      AI agents are already integrated into the government

      This is not a theoretical consideration for Estonia. The country has deployed AI chatbots in every school through collaborations with OpenAI and others, and operates Bürokratt, an expanding network of AI agents managing public services.

      Michal, who has an AI advisory council filled with tech entrepreneurs, recently created a “PM Cockpit” to monitor government objectives during a vibe-coding session using Anthropic’s Claude. Other nations are also progressing, from Ukraine’s Diia.AI to licensing pilot programs in Singapore.

      The unresolved issues

      Michal did not provide a start date or details on how liability would function if an agent with an independent ID makes an error. Those are the challenging questions that remain unanswered.

      Granting machines official identities within essential state systems poses a significant accountability and security risk. However, it also reframes the issue of agents beneficially: the aim is not to liberate the agents but to maintain control over them in a visible manner, as Europe hastens to establish its own regulations on AI.