Analysis
Policy
The EU AI Act's Second Act
The first wave of compliance was theater. The second wave is starting to bite — and not where most people expected.
By The Memo · Thursday, June 11, 2026 · 8 min read
When the EU AI Act's general-purpose model obligations went live, the consensus was that the rules would be either toothless or catastrophic, depending on which lobbyist you asked. Neither prediction has held up.
What's actually happening is more interesting: the Act is reshaping where models get evaluated, where weights get hosted, and which providers are willing to serve which European customers — and the effects are showing up in procurement contracts, not in headlines.
The training-compute threshold that triggers systemic-risk obligations was set as a bright-line rule. Bright-line rules in compute are unstable, because compute curves move quickly. We are already at the point where mid-tier open models trip the threshold under any reasonable reading, which is creating a steady drumbeat of legal-ambiguity questions the Commission has been slow to answer.
Subscribers only
Keep reading — it's free.
The Model Memo is a free daily newsletter. Drop your email to unlock the rest of this essay and get tomorrow's in your inbox. Always free, unsubscribe anytime.
Free daily newsletter. Unsubscribe anytime. No spam — ever.
Keep Reading
