Press Release July 31, 2026

CeSIA's three priorities make it into the French Parliament's AI report

Sénat

Senator Vanina Paoli-Gagin's report on AI alignment was delivered to Minister Anne Le Hénanff on July 30, concluding the mission entrusted by the Prime Minister. As the report itself puts it, alignment is not merely a competitive advantage but an existential and civilisational issue. The Hugging Face incident gave that urgency concrete form: OpenAI suspended training on the model involved and handed the analysis to external evaluators. Alignment is no longer a philosophical question; it is the condition for continuing to develop this technology with confidence.

CeSIA defended three priorities during the Senate hearing that informed the report:

  • Building independent capacity to evaluate and verify AI systems (priority 4). At present, companies set their own risk thresholds and determine on their own whether they meet them, an arrangement with no parallel in any other industry.
  • Giving France operational governance over alignment (priority 6), by resourcing INESIA to act at the pace AI capabilities actually move.
  • Preparing for resilience and pursuing an alignment diplomacy (priority 10). No alignment policy can rely on prevention alone, since AI systems remain insufficiently robust. The report calls for national and European exercises simulating loss of control (recommendation 28), accelerated cyber-defence capacity to preserve essential functions, and international coordination: a technical coordination and emergency-notification body modelled on the Financial Stability Board or the IAEA (recommendation 29), and a "coalition of the aligned" (recommendation 40). This is where France can exceed its budgetary weight, as a permanent member of the UN Security Council, host to the OECD and UNESCO in Paris, and holder of the G7 presidency.

Two French organisations are among the evaluators selected by the European AI Office for general-purpose AI models: CeSIA on manipulation risks, and SaferAI on CBRN risks. Four public operators are building national capacity alongside them: LNE, PEReN, ANSSI, and Inria.

CeSIA is pleased that its work contributed to this report and remains available to support implementation of this action plan.

The essentials of AI safety, every two weeks.

Get our selection of AI safety analysis and news in your inbox every two weeks.