French MPs Push Pre‑Approval Regime and Up to 7% Global Revenue Fines for General‑Purpose AI

On September 15, 2026, René Pilato together with 69 other deputies belonging to the La France insoumise (LFI) parliamentary group formally introduced Bill 3149 to the French National Assembly. The draft legislation is aimed at establishing a pre‑authorization regime for commercial general‑purpose artificial‑intelligence models as well as for high‑risk AI systems already identified in Annex III of the EU AI Act.

The proposal is presented explicitly as an opposition measure and has not yet been enacted into law. As of September 25, 2026, it has only been referred to the Assembly’s Economic Affairs Committee, meaning the text is still subject to debate, amendment and possible restructuring before any parliamentary vote can occur.

Key Features of the Draft Regulation

If the bill were to pass, it would create an 18‑member National Artificial Intelligence Authority, known by the acronym ANIA. This body would be charged with administering the new pre‑approval system, including the power to grant, suspend or withdraw authorisations for AI systems that fall within the scope defined by the legislation.

The draft defines a “substantial change” as any modification that materially alters a system’s compliance status or its risk profile. Whenever such a change occurs, the operator would be required to submit a fresh authorisation request, ensuring that ANIA reviews every significant update before the model can continue to be offered commercially.

Enforcement powers granted to ANIA would encompass orders to bring a system into compliance, the ability to impose daily penalties of up to €100,000, and the authority to suspend the operation of a system or to withdraw its authorisation altogether.

Financial Penalties and Reporting Obligations

Administrative fines under the bill could reach up to 7 % of a company’s worldwide annual turnover, aligning the sanction level with the most severe penalties in the EU’s digital regulatory framework. The calculation of the fine would be anchored to the fiscal year in which the breach is identified.

Beyond monetary sanctions, the draft mandates rapid incident reporting: any serious AI‑related incident must be reported to ANIA within 48 hours. Where the incident involves personal data, the responsible party must erase the data within seven calendar days of the report.

Human Oversight and Employment Decisions

The legislation expressly prohibits employment decisions that have legal or significant effects from being made solely on the output of an AI system. Such decisions must be accompanied by documented human review, ensuring that a qualified person validates the AI recommendation before it influences hiring, promotion or termination processes.

• Prior authorization required for commercial general‑purpose AI models
• Authorization renewal for any substantial compliance‑or‑risk change
• Creation of an 18‑member National Artificial Intelligence Authority (ANIA)
• Daily penalties up to €100,000 and fines up to 7 % of global revenue
• Mandatory incident reporting within 48 hours and data erasure within seven days

One unresolved issue concerns the potential overlap with the European Commission’s exclusive supervision of obligations for general‑purpose models under the EU AI Act. The French draft does not specify how national pre‑authorization would coexist with the EU‑wide framework, raising questions about possible legal conflicts and jurisdictional duplication.

For English‑speaking organisations that operate AI services in France, the draft signals a possible shift toward much tighter national oversight. Companies would need to prepare for a dual‑layer compliance process: first securing EU‑level conformity under the AI Act, then obtaining separate French pre‑authorization for any general‑purpose model or high‑risk system.

Implementation Timeline and Political Context

The bill’s progress will depend on the pace of deliberations within the Economic Affairs Committee and the broader political dynamics in the Assembly. Supporters argue that a national pre‑approval mechanism is necessary to protect citizens from emerging AI risks, while opponents warn that it could fragment the single market and create regulatory uncertainty for innovators.

While the proposal remains a draft, its existence urges multinational firms to begin mapping French regulatory expectations into their broader compliance strategies. The financial exposure could be significant, given that fines are calibrated to global turnover, and the requirement for documented human review of employment‑related AI decisions would necessitate adjustments to HR workflows and internal audit procedures.