On 7 May 2026 the European Parliament and EU member-state representatives reached political agreement on amending the AI Act (commonly called the “AI Omnibus”). According to public reporting, the amendment agrees to ban AI systems from generating deepfake pornography, sets a clear implementation timetable for the rules on high-risk AI systems and simplifies innovation-facing rules.
“Simplification” and “red lines” in parallel
The direction of this amendment can be described as two moves at once. One is subtraction: simplifying compliance burdens so that rules keep pace with innovation and companies do not spend excessive resources on procedural requirements. The other is addition: drawing clearer red lines around specific risks, deepfake pornography being one.
The two are not contradictory. Earlier criticism of the AI Act was split between “rules too heavy, implementation too slow” and “insufficient response to the risks of generated content”. This amendment answers both: loosening the innovation-facing parts and tightening around clearly harmful uses. Reading this dual track is the premise for judging where obligations go next. For an earlier walk-through of compliance pathways, see our EU AI Act compliance guide.
Banning deepfake pornography: a clear red line
The most closely watched item is the ban on AI systems generating deepfake pornography. It moves regulation that previously sat mainly at the “transparency” layer up to the “prohibition” layer.
The change has two implications. First, for content platforms, generating, hosting and distributing such content requires more proactive detection and handling mechanisms, rather than relying on complaints afterwards. Second, for companies providing generation capability, model and product safety strategies must cover this use — separating capability from clearly harmful scenarios through use restrictions, input and output filtering, and abuse monitoring.
For companies going global, the existence of a red line shifts the priority of risk assessment. Products involving image or video generation should first confirm whether their capability could be used to generate such content, then configure controls accordingly.
A timetable for high-risk rules
Another element is a clear implementation timetable for the rules on high-risk AI systems.
The value of a timetable is predictability. High-risk compliance often affects product architecture, testing processes and documentation, requiring a longer runway. In the past, ambiguity about when rules took effect and applied made it hard for companies to allocate resources. A clear timetable lets companies fold the compliance rhythm into the product roadmap.
A clear timetable is not a relaxation of obligations. The substantive requirements for high-risk scenarios — risk management, data governance, record-keeping and human oversight — do not disappear because of simplification. The sensible choice for companies is to work backwards from the timetable, not to strike compliance from the plan. The EU’s transparency and labelling duties are discussed in our EU AI Act transparency rules analysis.
Transparency duties: identifiable and clearly labelled
Running alongside the ban is the continuation of transparency duties. Two points are central: AI-generated or AI-modified content should be identifiable; and deepfake content must carry a prominent label.
Both have a direct effect on product flow. Identifiability means marking must be embedded at the content-production stage, and a prominent label means the notice to users must be clear and visible, not tucked away. For companies serving both the EU and UK markets, the two jurisdictions’ developments should sit on a single compliance map; see our multi-jurisdiction compliance guide and AI cross-border compliance under dual regulation.
What it means for companies going global
The signal from this amendment is that the EU is seeking a new balance between “innovation-friendly” and “risk-controllable”. For companies going global, the pragmatic response has three parts: reclassify products by risk to confirm whether they touch a red line or the high-risk list; build labelling and content-safety strategies into the product flow rather than retrofitting; and watch the subsequent official text and implementing rules, since room for change remains between political agreement and formal adoption.
For support on judging AI Act obligations, safety strategies for deepfake content and multi-jurisdiction compliance, see our EU AI Act service.
This article is based on public reporting on the political agreement reached by the European Parliament and EU member-state representatives on 7 May 2026 on amending the AI Act. It is for reference only and does not constitute legal advice. Specific compliance judgements should be based on the official text of the AI Act and official EU guidance.