A French expert commission handed President Emmanuel Macron a report on November 10, 2026 proposing 18 safeguards for users of generative AI. Protéger sans empêcher: nos vulnérabilités à l’ère de l’IA générative calls for independent evaluation and stronger protection of users. Its recommendations are proposals for public policy; the handover does not enact them or amend the EU AI Act.
The Commission on Generative AI and Vulnerabilities was convened through France’s Council for AI and Digital Technology, known as CIANum. According to the council and the Élysée’s handover account, its nine experts drew on 55 hearings, including contributions from public research and higher education.
Evaluation before market access and after release
The Élysée identifies prior multidisciplinary assessment as a central proposal and specifies its scope: consumer generative AI systems, rather than professional systems. That distinction matters for universities interpreting the announcement: it is not a newly imposed approval process for every research tool.
Four linked recommendations in the commission’s report set out the proposed evaluation framework:
- Recommendation 3: Make market entry conditional on assessment of whether consumer conversational AI exploits or aggravates user vulnerabilities.
- Recommendation 4: Require continued scrutiny after release, including periodic review, reassessment following substantial changes and a mechanism for alerts.
- Recommendation 5: Establish a binding framework giving authorities and accredited researchers access to usage, training and monetisation data.
- Recommendation 6: Organise interdisciplinary evaluation, strengthen the European AI Office’s role and provide lasting resources and broader powers for France’s national AI evaluation and safety institute, INESIA.
The report also proposes independent user panels and controlled access to models. Its evaluation agenda joins technical expertise with disciplines including law, sociology, psychology and economics.
What the proposals mean for university research offices
For research offices, the practical implication is an agenda to assess, rather than a new compliance deadline. Universities could contribute independent evidence through teams spanning technical and social research. An institutional review could ask which expertise, data access and safeguards such work would need.
The distinction between assessing a product before release and following its effects afterward also gives institutions a useful question for their own oversight: what evidence would justify revisiting an earlier assessment? CASRAI’s AI evaluation card entry provides related context for documenting evaluations.
AI literacy alongside institutional scrutiny
The council’s summary also prioritises broad education so people can understand AI’s uses and risks. It argues that responsibility for protection should not fall primarily on users or their families.
For universities, that combination suggests treating AI literacy and institutional evaluation as complementary tasks. Training can help students and staff question outputs and recognise limits, while research and oversight examine the systems themselves. The commission’s report offers a policy direction for that discussion, without establishing a new university implementation timetable.
Sources
- Élysée: handover of the expert commission’s report. November 10, 2026. Checked 2026-11-14.
- Élysée: ninth Paris Peace Forum, including the report handover. November 10, 2026. Checked 2026-11-14.
- CIANum: Protéger sans empêcher, report summary. Checked 2026-11-14.
- Commission on Generative AI and Vulnerabilities: full report. November 2026, recommendations 3–6 and 9. Checked 2026-11-14.
