The EU’s signature AI law just got pushed back. Not because it’s working well — but because it’s proving harder to implement than anyone thought.

The Delays

The European Parliament voted to delay compliance deadlines for high-risk AI systems from August 2026 to December 2027. That’s a 16-month extension. Companies making AI for sectors with existing safety rules (toys, medical devices) get until August 2028. The watermark requirement — forcing AI firms to label generated content — gets pushed to November 2026.

This is a significant softening. The EU AI Act was the world’s first comprehensive AI regulation, and it was supposed to start enforcing this year. Now companies get more breathing room. The official reason: businesses need more time to comply.

The real reason: the EU fumbled the implementation. As reported in Euractiv, the Commission missed its own deadlines to publish key guidance. The complexity of the rules caught up with the ambition.

The Nudify Ban

Here’s the part that actually matters: lawmakers also backed a ban on “nudify” apps — services that use AI to remove clothing from photos, creating non-consensual intimate images. The vote follows months of outrage over Grok’s image generator on X, which flooded the platform with AI-generated sexual deepfakes.

The language is vague. The ban “would not apply to AI systems with effective safety measures preventing users from creating such images.” That’s essentially saying: “ban the abuse, but we’re not sure how to do it technically.” It’s classic regulatory problems-in-search-of-solution language.

There are no specifics on enforcement. No details on what “effective safety measures” look like. Just a political signal that this behaviour is unacceptable.

What’s Next

Now comes the hard part. The European Parliament can’t simply change EU law on its own — it has to negotiate with the Council (ministers from all 27 member states) on the final text. That could take months. Or years.

This creates a strange situation: the law is on the books, but the rules are unsettled. Companies that wanted clarity get uncertainty. The regulatory framework that was supposed to give Europe a competitive advantage in “trustworthy AI” now looks like a work in perpetual progress.

The Bigger Picture

The EU is caught between its own ambition and its capacity to deliver. The AI Act was sold as Europe’s answer to American AI dominance and Chinese AI authoritarianism — a third way that balanced innovation with fundamental rights.

Instead, it’s become a cautionary tale. The US is rolling back state AI laws (more on that in the US section). The EU is still figuring out what it wants to be when it grows up.

The nude ban might actually happen — there’s political will. Everything else? Que sera, sera.