AI Governance Daily, 18 May 2026

Stop waiting for the legal-press summary. It's already late.

The European Commission published the draft Art. 50 transparency guidelines on 8 May. Lucilla Sioli, Director of the EU AI Office, shared the link on her own LinkedIn on Friday. That's the regulator handing you the operational playbook for 2 August 2026, in public, with a "tell us what you think" attached.

You have 16 days. Consultation closes 3 June.

For the last week I've been writing: don't guess what Art. 50 means in practice. Now there's nothing left to guess about. The draft is there. The submission form is there. SME exec + 90 minutes with a coffee = a primary-source read of what August looks like, plus the option to shape it.

If you do one governance thing this fortnight, do this.

🇪🇺 Europe & Regulation

The Art. 50 draft guidelines are live, and the AI Office wants your feedback. This is the operational tool for transparency. Informing people they're talking to an AI. Machine-readable marks on generated and manipulated content. Deployer duties on deepfakes, public-interest content, and emotion or biometric-category systems. Stakeholders explicitly invited, SMEs and citizens named. If you sell into the EU, this is the document your August 2 plan should be built on. Not a Reuters summary of it. The thing itself. Read

The submission page is the one bookmark that matters this fortnight. Read the draft. Find the bits that will cost you real money or break a real workflow. Write them up. Submit. Regulators read what's sent in. Silence = "the draft's fine as written." Read

🇺🇸 US

OpenAI gives the EU early access to a new cyber model. Anthropic still won't hand over Mythos to US federal procurement. Two labs, two strategies. OpenAI is playing the cooperation card on both sides of the Atlantic. Anthropic is drawing lines and eating the procurement hit. For governance leaders this isn't gossip, it's a signal about which vendors will and won't bend when a regulator asks for something awkward. Bake that into your vendor risk register. Read

NPR on the Trump pivot toward FDA-style AI oversight. A weekend analysis worth your time if you've been reading the administration as purely deregulatory. The mood music is shifting. Pre-market review language is back on the table. US ops teams should stop pretending federal AI rules aren't coming. Read

🧠 Voices worth 5 minutes

Azhar, Exponential View #574: inside Anthropic, AI pluralism, context-maxxing, Voltaire. Azhar's weekend essay collection. The Anthropic profile pairs nicely with the CNBC piece above, you get the strategy from the inside and the procurement consequence from the outside in the same morning. Read

⚠️ Watchlist

• Art. 50 transparency consultation closes 3 June 2026. 16 days. Read the draft, submit. This is your direct channel to the regulator.
• Art. 50 transparency obligations apply 2 August 2026 per the EC primary source. Plan to that date.
• EU Omnibus formal adoption still pending. If it lands, some dates slip. Plan as if it won't.
• Trump AI security executive order expected. A working group of tech executives and US officials is being designed. Watch the names.
• High-risk AI (Annex III) obligations apply 2 August 2026 on the current calendar.

---

One job this week: open the draft, mark it up, send it in. Your future August self will thank you. See you tomorrow.