AI Governance Daily, 26 May 2026

Yesterday I told you the EU operational machine is humming. Code of Practice landing, two consultations open, EDPB plenary Thursday.

Today Lawfare drops a quietly devastating piece on the other side of the Atlantic.

An AI executive order written to be maximally industry-friendly, one that literally disclaimed mandatory authority, still couldn't survive a handful of last-minute lobbying calls. That's the contrast: Brussels publishes 174-page rulebooks and runs public consultations. Washington governs by phone call.

If you're an SME exec watching both jurisdictions, you need to know which playbook is operating where. One is slow, transparent, and binding. The other is fast, opaque, and reversible by Tuesday afternoon.

🇪🇺 Europe & Regulation

• CNIL widens MR-001 and MR-003 for health research. The French DPA just extended its reference methodologies to cover foreign studies, paperless patient information, remote quality control, and access to identification data. If you're building or selling health-AI into France, your DPIA template just changed. = compliance work you thought was done is now partially undone. Read

🇺🇸 US

• AI Governance by Phone Call. Lawfare's piece is the cleanest articulation yet of what's actually happening in Washington. An EO drafted to give industry everything it wanted still got rewritten over the phone in the final hours. The lesson for any SME selling into US public sector AI: there is no stable rulebook to plan against. There's a frequency to tune into, and most of you don't have the lobbyist on retainer to hear it. Read

🏢 Enterprise & Operating Model

• AI in education, from the operators. Adeel Khan (MagicSchool AI) and Ryan Trattner (StudyFetch) on Scaling Laws. Education is Annex III high-risk territory under the AI Act, so this is not a fluff segment. Listen for how the operators talk about teacher trust, data flows, and what they think they can ship. Then ask whether your own high-risk use case is documented to that standard. Read

🧠 Voices worth 5 minutes

• Ethan Mollick: Choosing to Stay Human. As AI-generated output homogenises, the thing that distinguishes a brand, a memo, a board paper, is the part the model can't average toward the mean. Mollick on what stays human and how to keep your voice. Worth reading before your next all-hands. Read

⚠️ Watchlist

• Article 50 transparency consultation closes 3 June 2026 (8 days).
• High-risk classification consultation closes 23 June 2026 (28 days).
• Code of Practice on AI-generated content finalisation expected May/June 2026.
• EDPB 120th plenary on 28 May 2026 (2 days).
• Article 50 transparency obligations apply 2 August 2026.
• High-risk AI obligations apply 2 August 2026 on the current calendar.
• EU Omnibus formal adoption still pending.
• Trump AI security executive order still expected.

---

Two jurisdictions, two operating systems. Pick the one you can actually plan against, and stop pretending the other one will give you a memo. See you tomorrow.