Metsola warns MEPs not to rely on AI when drafting laws

European Parliament President Roberta Metsola has warned MEPs not to be overreliant on artificial intelligence in decision-making, as the Parliament prepares new guidelines for lawmakers on how to use it.

Speaking at an event on AI use in the Parliament on Wednesday, Metsola said: “Drafting an amendment with the help of AI cannot replace fundamentally the constructive debate with a colleague.” She added: “Doing legal research with the help of AI cannot ultimately replace the judgment and the critical thinking needed to make a tough decision.”

Metsola said there must be “democratic accountability” in lawmaking and argued that “where AI is used, people should know.”

“No matter how AI may be used in a parliament, the responsibility is still [with] those who are elected,” she said.

At the same event, Secretary-General Alessandro Chiocchetti said Parliament is building its own AI tool for sensitive work, hosted on its internal infrastructure.

As POLITICO reported in July, commercial AI models such as ChatGPT, Claude and Meta's Llama can still be used for less sensitive tasks.

Parliament officials have warned of a surge in AI-written amendments, including some that reference laws that do not exist. Chiocchetti cited the case of an MEP who turned to AI for legal advice and got a wrong answer. “Our jurists took two days just to prove ChatGPT wrong,” he said.

Chiocchetti said the administration will submit new guidelines on using AI to senior leaders “very soon.” Existing guidelines for staff date from 2024.

He said the new guidance boils down to three things: be open about using artificial intelligence, as the rules require; never put sensitive material into a public tool; and check everything before it leaves your desk, because it is your name on it.

Chiocchetti added that he wants to use AI to help sort through thousands of amendments. But the drafting and consideration of laws need to remain with MEPs, as “the risk is indeed that we delegate the political debate to a chatbot.”

“Every text we vote has to be one that people wrote, checked, and are willing to defend. Anyone reading a new regulation should be able to find out who decided and hold them responsible,” he said.