News
Brussels Delays AI Act Deadlines but Speeds Up AI Content Labeling
The European Parliament has approved the Omnibus VII package, pushing back some obligations for high-risk AI systems by up to a year and a half while accelerating rules on labeling AI-generated content and banning nudify-style apps.
On June 16, 2026, the European Parliament approved the Omnibus VII package, which changes the rollout schedule for the EU's AI Act. Companies preparing for the August deadline covering high-risk system rules get more time, but in exchange must move faster to label AI-generated content and implement a ban on apps that undress photos without consent.
The original AI Act timeline called for rules covering most AI systems, including high-risk ones, to take effect on August 2, 2026. High-risk systems include, among others, software used in recruitment, credit scoring, education, and healthcare. EU institutions reached a provisional agreement to change this timeline on May 7, 2026, and the European Parliament formally approved it six weeks later.
What Is Changing
The most significant change concerns standalone high-risk AI systems, meaning those that are not part of a larger product regulated under separate rules. Instead of August 2, 2026, companies now have until December 2, 2027, more than a year of extra time to adjust processes, documentation, and risk management systems. For high-risk systems embedded in other products, such as machinery or medical devices, the deadline has been pushed to August 2, 2028.
The European Commission had previously explained that it tied the effective date for high-risk system rules to the availability of harmonized standards, the technical benchmarks companies are meant to use to assess their systems' legal compliance. Some of these standards are still unfinished, which made it difficult for businesses to prepare for the original deadline.
Faster Than Planned
Not everything has been pushed back. The Omnibus VII package speeds up some of the rules on transparency for AI-generated content. The obligation to label such content, known as watermarking, is set to take effect as early as December 2, 2026, earlier than most of the rest of the AI Act's provisions. Alongside this, new bans have been introduced on generating privacy-violating content without consent, including nudify-style apps used to create undressed images of real people, as well as material depicting child sexual abuse.
This signals that Brussels treats deepfakes and content that violates personal intimacy as a more urgent threat than the risks tied to recruitment or scoring systems, even though the latter have long been the flagship example used to justify the entire AI Act. National regulatory sandboxes, programs letting companies test AI systems under regulatory supervision before full deployment, are to be established by August 2, 2027.
The Bill for Companies
For legal and compliance departments, this mainly means a shift in priorities, not relief. Companies that had put off classifying their AI systems, betting that the rules would take effect later anyway, now need to speed up specifically on content labeling, since that is the deadline that got shorter, not longer. At the same time, the extended timeline for high-risk systems does not remove the obligation to identify them, since risk classification remains a requirement independent of the timeline for full obligations.
From a business perspective, delaying the deadlines for changes to AI regulation is meant to give companies a better chance to prepare their processes and resources - Aleksander Zieliński, Sandra Winiecka, DSK Kancelaria
The Polish Perspective
In Poland, the problem runs deeper than deadlines: some public offices and institutions still aren't sure whether they even use systems that the AI Act would classify as artificial intelligence. The extra time to adjust high-risk systems could therefore, paradoxically, help public administration bodies that are only now learning to inventory their own tools before they can even begin classifying them by risk.
For Polish tech companies and integrators deploying AI systems for business clients, the key thing to track will be which specific systems fall into the high-risk category, since that classification determines whether they benefit from the extension to December 2027 or need to move faster because of the AI-generated content labeling obligations that take effect as early as the end of 2026.
The text of the Omnibus VII package still requires formal approval by the Council of the European Union and publication in the EU's Official Journal before the changes become legally binding. That process is expected to conclude before August 2, 2026, the date on which the rules for high-risk systems were originally set to take effect.
Sources: AI Act after Omnibus VII. New timeline for high-risk AI system requirements (gazetaprawna.pl), AI Act with a new timeline - not everything can wait (rp.pl), Omnibus VII package - new timeline for AI Act implementation (kancelariamacura.pl)
