Roberta Metsola, President of the European Parliament (pictured), has warned MEPs not to use artificial intelligence as a substitute for the legislative process, Politico reports. During a meeting on AI use in Parliament, she said elected representatives must remain personally responsible for all legislative decisions.
According to Metsola, AI can help draft amendments. However, the technology cannot replace debate between parliamentarians or the critical thinking needed for complex legal decisions.
To safeguard democratic transparency, Metsola wants Parliament to communicate openly about its use of AI. This would allow citizens to hold their representatives to account.
That concern is growing as parliamentary staff increasingly encounter AI-written amendments. Some even refer to laws that do not exist. Secretary-General Alessandro Chiocchetti highlighted the risks with a concrete example. An MEP received incorrect legal advice from an AI tool. Legal experts then needed two days to demonstrate the error.
New guidelines for AI use
Chiocchetti believes MEPs must remain responsible for drafting and assessing laws. AI can, however, help organise large numbers of amendments.
The parliamentary administration is therefore working on new guidelines. These are based on three principles: transparency about AI use, not uploading confidential documents to public platforms, and thoroughly checking all AI output.
Different rules currently apply to staff and elected members of Parliament. Since 2024, staff have been required to disclose when they use AI for their work. MEPs are not yet subject to that requirement for speeches or amendments.
To introduce binding rules for them as well, Parliament may need to amend its Rules of Procedure. The Committee on Constitutional Affairs would then have to consider that change.
Alongside the new guidelines, Parliament is developing its own AI platform for sensitive information: the EPGenAI Hub. This application runs on Parliament’s internal infrastructure and is intended to enable the secure processing of confidential data.
Commercial applications such as ChatGPT, Anthropic’s Claude and Meta’s Llama will remain available for less sensitive tasks. Parliament also already takes a cautious approach to hardware. For example, the institution previously disabled integrated AI features on tablets to protect data privacy.
Business AM
Ltd
