Estonia plan for AI agent codes is dangerous, experts tell

Estonia plans a law that lets persons register an agent code for an AI agent that does tasks for them. Liina Vahtras, the minister of economic affairs of Estonia, tells that the plan is to show who uses each agent. Cybernetica sent 11 pages of feedback and tells that the bill is dangerous. Only the Government Office approved the bill before the deadline.

How it works

An AI agent gets a task and finds a procedure to do it. A person can register one agent code in one database. Then each person can find which agent works for whom. The first plan let companies register many agents. At this time only persons, not companies, can, with one agent for each person.

The numbers

The source gives no test results. It gives 11 pages of feedback and one approval. The bill must go to the Riigikogu, the parliament of Estonia, at the start of November.

What is not known

The software for the plan is not known. Arne Ansper of Cybernetica tells that registration gives no value without strong sign-in at the level of ID cards. Ats Albre of an Estonia group of companies tells that the person must stay responsible. Vahtras tells that the plan has no duties and the government can stop it.

Insight: Registration is becoming a gate for agents

Salesforce will make its customers register external agents that use its APIs (software connections) from November 2026. Each agent gets a small set of permissions. Estonia plans a registry too, and critics tell that a code with no strong sign-in proves no value. It seems that the maker of the sign-in tool, not of the registry, will make the rules. It is possible that you must register agents in more than one system.