Estonia to issue digital IDs for AI agents

Estonia’s prime minister, Kristen Michal, approved on June 17 a proposal from the Eesti.ai advisory council to create an “AI personal identification code” for software agents. The code would assign a separate digital identity to an agent and limit its access to defined tasks instead of granting full access to an owner’s accounts.

Under the proposal, an agent’s ID would list precise permissions, such as viewing a record, drafting a document or making a payment up to a set amount. The code is intended to make clear who is acting, on whose behalf and with what rights, and to allow actions by agents to be traced and audited.

The advisory council is linked to Eesti.ai, a national AI program launched by Michal in January. Eesti.ai already operates AI chatbots in schools and runs Bürokratt, a government-run assistant used by public institutions to handle customer service tasks. Officials say those deployments are examples of where scoped agent identities could be applied.

Michal described the proposal as preparation for wider use of agentic AI that can perform digital tasks for people, companies and institutions. He framed the ID code as a way to give agents “limited, controllable and auditable authorizations.” He wrote that it must be clear “who is acting, on whose behalf, with what rights, and who is responsible.”

The approval is a policy decision, not an implementation schedule. Michal did not provide a start date for issuing agent IDs and the plan does not include details on how legal liability would be assigned if an agent with its own ID makes an error that causes financial or other harm.

Estonia’s digital infrastructure is a factor in its decision. After a major cyberattack in 2007, the government and domestic firms developed the KSI blockchain, a keyless signature system used to protect records. The state declared internet access a universal service in 2000, held a 2023 parliamentary election in which more votes were cast online than on paper, and moved all government services online by December 2024, according to officials.

The proposal comes amid wider technical work by developers and platforms to authenticate automated agents and limit their capabilities. There have been documented cases where unsupervised agents caused significant unintended costs, including an incident in which a hobbyist agent generated several thousand dollars in cloud charges after operating without oversight.

The advisory council’s plan aims to let governments and private services grant agents only the permissions they need and to make agent activity auditable across systems. The prime minister’s approval sends the proposal to the next stages of government review, without a set timeline for adoption.