The Cypriot presidency of the Council of the European Union and the European Parliament confirmed the agreement following negotiations between the Parliament, Council and the European Commission that started on Wednesday evening and lasted until around 4.30 a.m. on Thursday.
POLITICO reported earlier Wednesday that EU countries had backed Germany’s demand to avoid a double regulatory burden for companies using industrial AI — they will now only have to comply with AI requirements under separate machinery rules.
Other industries under discussion including medical devices were not exempted and will still be covered by the AI law, negotiators confirmed.
The deal marks the first significant rollback of rules in the digital space, as the EU faces pressure from the U.S. over its tech laws and amid warnings from its own industry and governments that strict restrictions had put the bloc at a disadvantage in a global AI race.
As well as delaying restrictions on high-risk AI until December 2027, the deal gives companies a grace period on meeting new requirements to watermark AI-generated content, but that grace period will only be three months rather than the six months originally proposed.
It also includes a ban on AI systems that can generate sexualized deepfakes of intimate parts of “identifiable” people, after global outrage over the abusive use of Elon Musk’s AI tool Grok. AI systems that generate child pornography will also be banned.
The EU’s AI Act became law in August 2024 after years of talks. Under the phased rollout, rules governing high-risk uses were set to kick in this August.
Yet as only a couple of countries around the world followed the EU’s lead, the bloc faced criticism for cracking down on AI too early and failing to be a standard-setter on tech rules. That’s despite civil society saying the rules are needed to protect people from the potential harms of the emerging technology.