AI Governance Daily, 20 May 2026

Three days ago I told you the Sioli transparency consultation was the regulator's homework for SMEs.

Yesterday Lawfare answered how much enforcement muscle sits behind the AI Office.

Today Brussels dropped the OTHER consultation everyone has been waiting for: the draft guidelines for classifying high-risk AI systems. 174 pages, 3 documents, 3 months late, but here they are.

That's two open consultations now. Both close in the next 34 days. Both shape the AI you ship next year. If your system touches hiring, lending, education, healthcare, critical infra, law enforcement, migration or justice, this is THE document. SME execs who let this window slide shut = the ones writing apologies in October.

Let's get into it.

🇪🇺 Europe & Regulation

• Commission publishes draft guidelines on classifying high-risk AI systems. 174 pages across 3 papers. This is the rulebook that decides whether your AI lands in the heavy-compliance bucket or skates past it. Annex III's 8 areas (biometrics, critical infra, education, employment, essential services, law enforcement, migration, justice). Classification by intended purpose, not by how much human oversight you bolted on after the fact. Article 6(3) exemptions and how to actually claim them. Split architectures and agentic AI treated as one system, which is the right call and also the expensive one. Read it, mark it up, file by 23 June. Read

• IAPP: EC delivers high-risk guidelines after delays. 3 months late, but landed. If you need the in-house-counsel summary before you crack the 174 pages, start here. Read

• AI Act Newsletter #102: pressure builds over Anthropic's Mythos. Digital Omnibus deal struck, transparency guidelines out, agentic AI readiness questioned. The week-in-Brussels round-up I always send to colleagues who don't want to live in Substack. Read

🇺🇸 US

• Colorado revises its AI Act. Governor Polis signed SB 189 on 15 May after two years of negotiation. The US state-level compliance map just shifted again, and if you sell into Colorado you can't pretend this is somebody else's problem. Read

• FTC begins enforcing the TAKE IT DOWN Act. New FTC complaint portal for platforms failing to remove non-consensual intimate images on request. Last week was warning letters. This week is enforcement infrastructure = the regulator means it. Read

🏢 Enterprise & Operating Model

• Anthropic + KPMG strategic alliance. Claude rolled across KPMG's 276,000 people. Big-4 AI vendor consolidation is happening in public now, and your procurement team should be reading the contract terms, not the press release. Read

• OpenAI launches OpenAI for Singapore. Multi-year partnership covering deployment, local talent and public services. Singapore keeps quietly positioning itself as the practical AI governance hub while everyone else writes white papers. Read

• OpenAI: advancing content provenance. Content Credentials, SynthID and a verification tool for AI-generated media. Direct relevance for the Article 50(2) watermarking obligation that lands 2 August. If you're a deployer, this is the supplier side of your compliance story. Read

🧠 Voices worth 5 minutes

• Lawfare: The AI Race Isn't Real. Why the "race with China" framing isn't a race and isn't worth running. Print this and slide it across the table the next time your board uses "we're falling behind" to justify skipping governance. Read

• MIT Sloan: Companies Don't Have to Slash Jobs Because of AI. The counter-narrative to the layoff drumbeat your CEO has been marinating in. Evidence, not vibes. Read

⚠️ Watchlist

• High-risk classification consultation closes 23 June 2026 (34 days)
• Article 50 transparency consultation closes 3 June 2026 (14 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, 34 days, one August deadline that doesn't care if you read the PDFs. Pick one. File something. See you tomorrow.