AI Governance Daily, 25 May 2026

Friday I told you the EU AI Act stopped being theoretical.

Today the calendar tightens.

Art. 50 transparency consultation closes in 9 days. The voluntary Code of Practice on marking AI-generated content, the practical compliance tool for Art. 50(2), finalises in the next 4-5 weeks. Two consultations and one Code of Practice, all landing inside June.

Slow news Monday. Fast operational week ahead. Quiet inbox = false comfort. If your compliance plan still says "we'll look at watermarking in Q3," you've already lost the window.

🇪🇺 Europe & Regulation

• The Code of Practice is the cheat sheet, not the homework. Voluntary route to Art. 50(2) compliance. The watermarking and detection obligation lands 2 August 2026. If you sign the Code, you get presumed conformity. If you don't, you're writing your own compliance memo from scratch and defending it to a regulator. Pick your pain. Read

• Second draft is out, final is weeks away. Published 5 March 2026, shaped by hundreds of stakeholders. This is the version your legal and product teams should be red-lining now, not when the final drops. Governance leaders = read it this week, map your gen-AI surfaces against it, then brief your exec. Read

• National security governance starts before the courtroom. Lawfare reframes democratic defence as an operational discipline, not a legal one. Useful lens if you're building AI policy for regulated, defence-adjacent or critical-infrastructure clients. Read

🏢 Enterprise & Operating Model

• Ben Evans on AI job exposure: stop guessing, start modelling. Which jobs, which companies, which industries. This is the workforce-planning artefact your HR director needs before they touch Annex III hiring obligations. AI governance is not just model risk. It's people risk, and the Act treats HR systems as high-risk for a reason. Read

• The ghost token: why your AI bill keeps climbing. Azhar on token economics. Per-token cost falls, total bill rises, because usage explodes and one task triggers many model calls. If your CFO is asking "why is the AI line item up 40% when prices dropped," send them this. = unit economics not equal to total economics. Read

🧠 Voices worth 5 minutes

• Exponential View #575: math breakthrough, creative limits. Azhar's weekend essay. "Humans were not put on this earth to maintain Excel models." The capability frontier moved on quantitative reasoning. The creativity frontier didn't. Read it as a strategy prompt: where in your operating model are you still paying humans to be calculators? Read

⚠️ Watchlist

• Article 50 transparency consultation closes 3 June 2026 (9 days)
• High-risk classification consultation closes 23 June 2026 (29 days)
• Code of Practice on AI-generated content finalisation expected May/June 2026
• EDPB 120th plenary on 28 May 2026 (3 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 consultations, one Code, one plenary, all inside the next four weeks. If you're treating this as a watching brief, you're already behind.