Brussels did it again.

After months of trilogue haggling, the European Parliament, Council and Commission struck political agreement this week on the long-delayed AI Liability Directive. It's a companion to the AI Act. And it's going to make suing AI developers significantly easier. Globally.

The big change is technical but it matters: a rebuttable presumption of causality. If a claimant can show an AI system didn't comply with its AI Act obligations, courts will presume that non-compliance caused the harm — unless the defendant proves otherwise. That flips the traditional burden of proof in EU tort law on its head.

The directive also forces wide disclosure. Claimants can get courts to order defendants to produce technical documentation, training data descriptions, and risk management records. Stuff AI companies have treated as their crown jewels.

Reactions split predictably. Consumer groups love it. Industry lobbyists — both US hyperscalers and European startups — warn the compliance burden will slow product launches versus the US and Asia. One Brussels lobbyist quipped that the EU is becoming the place where AI gets regulated, not built. Cute. Not entirely wrong.

Member states have two years to transpose the directive. The earliest real claims should land in 2028. But here's the practical bit: every non-EU AI firm that serves European users needs to review liability exposure well before then. Roughly 300 global AI companies are directly affected by current estimates, and many are already beefing up their EU compliance teams. If you're running an AI product with European users, your lawyer's time is about to get expensive.