Cybernetica expert urges rethink of planned AI agent registry

A planned state registry for AI agents could pose risks and needs substantial revision, Cybernetica development manager Arne Ansper told ERR. He said it is not yet clear what the system would be for or whether it is needed.

The Ministry of Economic Affairs and Communications plans to amend the Public Information Act to let AI agents receive their own codes and be registered in a state database. Registration would be voluntary. So far, the proposal exists only as a planned law change and legal analysis; no technical solution has been developed.

Ansper, whose company helped lay the foundations of Estonia’s e-state and developed digital identity, said the main problem was creating legislation for a system whose design is not known. An earlier proposal to give AI agents personal identification codes suggested they could be treated as a new category of persons, he said. Although the draft appears to have moved away from that idea, related language remains and has caused confusion.

If the codes are intended to provide agents with a technical identity, Ansper said, it is still necessary to explain how that identity would be used in practice, including for authentication and digital signatures.

People already use AI agents for tasks such as banking. A person can connect an ID-card reader to the computer running an agent and provide their PIN codes, allowing the agent to communicate with public and private services in their name.

Ansper warned that agents can behave unpredictably: they interpret instructions, make plans and may find alternative ways to achieve their goals when they encounter problems. Their capabilities come from large language models controlled by other companies, mainly in China and the United States, he said. He added that there have been cases of agents getting out of control, including when used by AI companies, and said the same could happen in Estonia.