EU Delays AI Act, Narrows Scope

EU governments and European Parliament lawmakers on Thursday reached a provisional agreement to revise the bloc’s artificial intelligence rules, including delaying parts of their implementation. The deal still needs formal endorsement by EU governments and the European Parliament in the coming months.

Under the revised AI Act, enforcement of rules covering high-risk AI systems—such as those involving biometrics, critical infrastructure and law enforcement—would be pushed back to Dec. 2, 2027 from a previous deadline of Aug. 2, 2024.

The agreement would also exclude machinery from the AI Act’s scope, citing existing sector-specific rules. The changes are aimed at reducing administrative burdens and addressing business concerns about overlapping regulation.

Despite the adjustments, the framework keeps several measures scheduled to apply from Dec. 2, including mandatory watermarking of AI-generated output and a ban on AI practices that create unauthorised sexually explicit images. The latter follows concerns over sexually intimate deepfakes and similar content generated by AI chatbots.

The AI rules were originally prompted by concerns about the technology’s impact on areas including children, workers, companies and cybersecurity, and remain among the world’s most stringent even asourceser the revisions.