On June 26, 2025, two visions of the future of artificial intelligence came into sharp focus – and they couldn’t have been more different.
➡️ In Brussels, 46 major companies – including Google, Meta, Amazon, Microsoft, Mistral AI, and others – signed a letter calling for a two-year moratorium on the enforcement of the AI Act, fearing that regulation could stifle innovation.
➡️ In Paris, CNIL, ANSSI, PEReN, and the IPoP project from the PEPR Cybersecurity initiative launched PANAME, a project to audit the privacy compliance of AI models, fully aligned with the GDPR and the upcoming European AI Regulation.
📍 Same day → two frontlines, two philosophies.
🇫🇷 PANAME: Regulation through engineering
PANAME aims to equip GDPR compliance by developing a software library to test whether an AI model retains personal data.
- It will be partially open source;
- Reusable by auditors, companies, and authorities;
- And based on a unified, reproducible, and verifiable methodology.
This project follows a clear principle: don’t wait for incidents to happen before addressing risk.
⚖️ A challenge at the crossroads of law, tech… and sovereignty
➤ While Big Tech asks for more time, French regulators are building tools.
➤ Where some fear complexity, others confront on-the-ground realities.
For legal professionals, the stakes are immediate:
- How can we prove that a model does not retain sensitive data?
- How can we prepare for future audits?
- How can we manage algorithmic risk while staying competitive?
💡 PANAME could become to AI what crash tests are to the automotive industry: a recognized, enforceable, and reproducible verification protocol… Time will tell.
In the meantime, on June 26, two AI approaches clashed:
The United States, hitting the brakes on regulation to preserve speed, vs. Europe, equipping regulation to preserve trust.
In the end, who will prevail? Those who innovate without a safety net, or those who innovate with discipline?
Source : https://cnil.fr/fr/paname-un-partenariat-pour-laudit-de-la-confidentialite-des-modeles-dia
