Let’s be honest: the EU AI Act was always more ambitious than practical. And now Brussels has essentially admitted it.

After two failed trilogues — the closed-door negotiations between Parliament, Council, and the Commission — EU lawmakers finally landed a compromise deal that does three significant things:

First, the high-risk AI compliance deadline gets pushed from August 2026 to December 2027. That’s a eighteen-month stay of execution for companies that were staring down costly conformity assessments.

Second, smaller AI firms get relief from the worst of the paperwork burden. The original Act’s documentation requirements were designed for large enterprises with legal departments. Startups building AI products were looking at compliance costs that could have shuttered them outright.

Third — and this one is more specific — nudification apps are now explicitly banned across the EU. These apps used AI to remove clothing from photos, typically targeting women. The Act wasn’t originally designed to handle this specific abuse case, so they wrote it in directly.

Why Now?

The timing isn’t coincidental. The August 2 enforcement date was always going to be chaotic. Member states still hadn’t set up their national market surveillance authorities. The Scientific Panel — all sixty of them — was only just convened. And the practical reality of enforcing a regulation that touches everything from recruitment software to medical devices was hitting policymakers like a wall.

Lobbying helped too, obviously. The tech industry spent two years arguing that the compliance timeline was unrealistically aggressive. They were right, even if their motives were self-interest.

The Innovation Question

The Parliament Magazine article on why the EU rewrote its landmark law gets at the real tension: does regulation slow innovation? The EU’s answer in 2024 was “we don’t care, we’re doing it anyway.” Their answer in 2026 is beginning to look like “actually, yes, it might, so let’s ease off.”

Whether that’s the right call depends on your priors. If you think AI risks are serious and underweighted by markets, gutting the compliance regime early looks reckless. If you think the original Act was regulatory overreach that would have kneecapped European AI competitiveness, this looks like sensible course correction.

I’m not sure it’s either. The EU’s track record on tech regulation suggests they struggle to get the balance right in either direction — too light on platform power, too heavy on innovation. This deal feels like the latter finally yielding.

What Comes Next

The deal still needs formal approval from Parliament and Council. That’s expected to be procedural rather than contentious — the hardest negotiations already happened. By late 2026, the amended Act will be law, and the countdown to December 2027 will begin in earnest.

For AI companies building in Europe: your compliance burden is lighter than it was going to be. For everyone else: the debate about what the AI Act should actually accomplish is far from over.

The nudification ban, incidentally, will likely survive the final vote with bipartisan support. Sometimes regulation works exactly as intended.