European Parliament and Council negotiators reached a provisional agreement early Thursday on changes to parts of the European Union’s Artificial Intelligence Act as part of the bloc’s digital omnibus package.
The agreement delays some parts of the AI Act to give more time for standards and support measures needed to help apply the rules properly.
Under the deal, rules for high-risk AI systems used in areas such as biometrics, critical infrastructure, education, employment, law enforcement and border management will start applying from 2 December 2027.
Rules for AI systems used as safety components under EU safety and market surveillance laws will apply from 2 August 2028.
The agreement also delays rules requiring AI-generated content to include watermarks until 2 December 2026. Watermarks help identify and trace AI-generated material.
Parliament and Council also agreed to ban AI systems used to create child sexual abuse material or sexually explicit content involving an identifiable person without their consent.
The ban applies to companies placing such AI systems on the EU market, companies failing to include reasonable safeguards, and people using the systems to create such content.
The rules cover images, videos and audio. Companies will have until 2 December 2026 to comply.
Negotiators also agreed on measures aimed at reducing overlapping rules and simplifying enforcement.
Under the agreement, AI systems used in machinery products would only need to follow sector-specific safety rules instead of both the AI Act and sectoral legislation, while still maintaining health and safety protections.
The updated rules also narrow the definition of “safety component,” meaning products using AI only to assist users or improve performance would not automatically face high-risk obligations if failures do not create health or safety risks.
The agreement also allows personal data to be processed when strictly necessary to detect and correct bias in AI systems, provided safeguards are in place.
Exemptions available to small and medium-sized businesses under certain rules would also be extended to small mid-cap companies to support growth.
In addition, the agreement simplifies enforcement of some general-purpose AI systems through the EU’s AI Office.
Arba Kokalari (EPP,SE) said the agreement would make the AI rules “more workable in practice,” remove overlaps and support innovation and startups in Europe.
Michael McNamara (Renew, IE) welcomed the ban on “nudification apps” and AI systems used to create child sexual abuse material, saying the agreement would help protect fundamental rights and human dignity.
The provisional agreement still needs formal approval from both the European Parliament and the Council before becoming law.
EU lawmakers aim to adopt the legislation before 2 August 2026, when current rules on high-risk AI systems are due to begin.
