After nine hours of closed-door negotiations on Thursday, EU countries and European Parliament lawmakers agreed to water down the bloc's landmark AI Act. The deal delays enforcement of rules on high-risk AI systems by 18 months, excludes entire industries from coverage, and pushes back deadlines that were supposed to hit this August.

Let me say that again: the EU just voluntarily weakened the only comprehensive AI law on earth. The one they spent three years building. The one they told the world proved Europe could lead on tech regulation.

Here is what changed. Rules on high-risk AI systems, covering biometrics, critical infrastructure, and law enforcement, have been pushed from August 2, 2026 to December 2, 2027. Machinery is now fully excluded from the AI Act because it is "already subject to sectoral rules." And the whole package is framed as "simplification" to help European companies compete with American and Asian rivals.

Cyprus's deputy minister for European affairs, Marilena Raouna, framed it as a win: "Today's agreement significantly supports our companies by reducing recurring administrative costs." Translation: the rules were too hard, so we made them easier.

The Grok Exception

Not everything got weaker. The deal includes a new ban on AI that creates unauthorized sexually explicit images, which is a direct response to Elon Musk's xAI chatbot Grok generating non-consensual deepfakes on X. That ban takes effect December 2. Mandatory watermarking of AI-generated content also kicks in on the same date.

Dutch lawmaker Kim van Sparrentak called it a victory: "By the end of this year, everyone, but especially women and girls will be safe from horrific nudifier apps being widely available on the EU market." It is telling that the one area where the EU got tougher involves Musk's platform. The political will to regulate AI apparently depends on who is doing the damage.

What This Actually Means

The timing is brutal. This deal lands on the same day the Trump administration is actively considering pre-release safety testing for AI models. While Washington pivots toward oversight, Brussels is retreating from it. The two largest regulatory powers on earth are moving in opposite directions at the same time.

The EU officials insist the AI Act remains "the strictest in the world" even after the changes. That is technically true. But being the strictest law that keeps getting weaker is not the flex they think it is. The message to Big Tech is unmistakable: complain loudly enough about competitive disadvantage, and Europe will bend.

The tentative agreement still needs formal endorsement from EU governments and Parliament. But provisional deals at this stage almost never collapse. This is the new EU AI Act. Lighter, later, and a lot more friendly to the companies it was supposed to regulate.