AI agents will not get Estonian ID codes. The person who authorised them stays liable.

In June the prime minister floated giving AI agents their own personal identification codes. In September that plan was dropped. Anni Lehari, who leads the Aruait project, described ID codes for agents as a risk to the digital trust framework Estonia has spent thirty years building. The identity layer only works because an ID code means a person.

The replacement model is more familiar than the original. An agent is traceable to the human who authorised it, and authority flows through the agent the same way it flows through a delegation inside a company. The agent acts. The person who signed it over remains liable. Authorisations are revocable.

Why this is the right answer for companies too

Nothing about the alternative was ever attractive once you worked it through. An agent with its own legal identity is an agent you cannot sue, cannot fire and cannot ask to explain itself. Every liability question would have ended in a shrug.

The delegation model gives you something you can actually operate. It also hands you a question to answer internally, because the state has now made the structure explicit: for every agent running in your company, who authorised it, and does that person know they are on the hook for what it does?

Behind it, the plumbing

The Information System Authority is building central security and authentication standards covering roughly 800 public services, so agents acting on a person’s behalf can be recognised and stopped consistently rather than service by service. Aruait is a one million euro project running through 2028, covering the legal, data protection and cybersecurity side of the same problem.

Companies that automate dealings with public services should watch this rather than build around it. An authorisation standard you can plug into is worth more than a workaround you have to maintain.

What to put in place now

• ✳Keep a list of every agent in production, with the person who authorised it named on each line
• ✳Tell those people, in writing, what the agent is allowed to do on their authority
• ✳Set an expiry date on each authorisation so dormant agents lapse instead of lingering
• ✳Check that you can revoke one in minutes, not after a conversation with whoever built it

The useful part of this decision is that it removes a distraction. There is no new kind of legal person coming. There is a named human behind every agent, which is how it already worked in practice.