Human oversight and content creators always remaining responsible for content: these are the AI guidelines for MPs

Artificial intelligence is a useful tool for drafting legislation, but it can never replace human judgement. The guidelines for MPs, which were submitted on 30 July for consideration by the Supervisory Committee on Parliamentary Documentation at the Chamber of Deputies – chaired by Vice-President of the Chamber Anna Ascani (PD). Final adoption is scheduled for September.

This is a 33-page document aimed at safeguarding organisational and IT security, previewed in the print edition of *Il Sole 24 Ore* and drawn up in line with the European and national regulatory framework and the key principles underpinning the work of the legislature. ‘In the parliamentary context,’ the introduction states, ‘artificial intelligence cannot be regarded as a mere productivity tool. Its use, in fact, falls within a sphere characterised by constitutional responsibilities, the public significance of legislative acts, the guarantee of pluralism, the accuracy of information and the protection of fundamental rights.’

A preliminary remark is in order: these recommendations do not replace the security policies or the regulations on the protection of personal data established by the competent bodies of the Chamber of Deputies; rather, they provide general criteria and rules of conduct. Four fundamental principles have been set out in writing. The first is to always maintain human oversight (human in the loop), viewing AI as a support tool rather than a replacement for human judgement. Any content generated, summarised or reworked must therefore be verified before use, particularly when it is intended for inclusion in public or official documents. Responsibility for the content remains, in all cases, with the party ordering its use.

The second piece of advice is to be careful about what you share. Personal data, confidential documents, drafts of documents not yet submitted, or other sensitive information must not be entered into public artificial intelligence tools or, in any case, made available by the administration. The data entered could, in fact, be stored, reused for model training or made accessible to third parties.

The third point may seem obvious, but it is not. Members of Parliament are advised to always check their sources, as AI systems can generate information that is plausible but inaccurate or lacks adequate corroboration in the sources: these are known as ‘hallucinations’. Relevant information must therefore always be checked, and care must be taken to ensure that the generated content does not infringe intellectual property rights or other rights.