News report: the EU's AI omnibus reaches a deal, resetting the high-risk deadlines and adding a nudifier ban

This documentary photograph of the European Parliament's Brussels entrance fits the story because the key move is legislative bargaining among EU institutions: negotiators have agreed to delay parts of the AI Act while adding a new ban on AI nudifier systems.[6]

As of 2026-05-07T09:03:45Z (UTC), the European Union has a political deal, not yet a final law, on its AI simplification package. The agreement reached between Parliament and Council would push the main high-risk AI Act obligations off their current 2026 start date, set a new earlier date for watermarking rules on AI-generated content, and add an explicit EU ban on AI nudifier systems that create non-consensual intimate content.[1][2]

The most important point is the sequencing. Brussels has not abandoned the AI Act's risk-based structure. It has chosen to slow the most operationally difficult parts until the standards, guidance, and support measures are more likely to exist in usable form.[1][2][3][4] For companies, regulators, and procurement teams, that is a real schedule change. For policy watchers, it is also a clue about what worried lawmakers most: not the existence of the rules, but the risk that the toughest obligations would arrive before the compliance machinery around them was ready.

Image context: the cover uses a real photograph of the European Parliament building in Brussels because this story turns on co-legislators rewriting the implementation calendar. A generic AI illustration would misstate the event; the live development is institutional and procedural.[6]

Fact file

Status now: Parliament and Council negotiators reached a provisional political agreement on the AI omnibus; the text still needs formal adoption by both institutions before it becomes law.[1][2]
Main deadline reset: obligations for stand-alone high-risk AI systems would apply from 2 December 2027.[1][2][3][4]
Product-embedded systems: high-risk AI systems used as safety components in regulated products would apply from 2 August 2028.[1][2][3][4]
Watermarking change: obligations for AI-generated content watermarking would apply from 2 December 2026 under the deal, earlier than the 2 February 2027 date in the Commission proposal and later than the Parliament committee's earlier 2 November 2026 preference.[2][4]
New ban: the deal adds a prohibition on AI systems used to create non-consensual intimate content or AI-assisted child sexual abuse material.[2][3][4]
Immediate political objective: Parliament, Council, and Commission all frame the rewrite as a way to reduce legal uncertainty and make the AI Act workable before the current 2 August 2026 high-risk clock arrives.[1][2][3]

What changed on May 7

The Commission's May 7 statement is short, but it makes the core decision plain. The agreement keeps the AI Act in place while moving the application dates for high-risk systems to December 2027 and August 2028, depending on the category.[1] Parliament's readout fills in the operative detail. It says the deal also postpones the watermarking rules to 2 December 2026, preserves the ban Parliament had pushed for on so-called nudifier systems, and still needs to be formally adopted before the law can enter into force.[2]

That formal-adoption point matters. The current law still points to 2 August 2026 as the start date for major high-risk obligations. The provisional deal changes that only if the institutions finish the legislative process in time.[2] Parliament's negotiators explicitly said they intend to complete adoption before that August deadline.[2]

The compromise also shows where the institutions converged and where they split the difference. In March, the Council backed delaying high-risk obligations, fixing the new dates at 2 December 2027 and 2 August 2028, while also adding a ban on non-consensual intimate-content generation and reinstating some database-registration obligations for systems claiming exemption from the high-risk label.[3] Parliament's March committee position matched the new high-risk dates and pushed for a nudifier ban too, but it had argued for an earlier 2 November 2026 watermarking date.[4] The May 7 deal keeps the Parliament-backed ban, keeps the shared high-risk dates, and lands watermarking at a middle point: 2 December 2026.[2][4]

Why the delay is the real story

The headline temptation is to read this as either "Europe retreats on AI rules" or "Europe tightens AI safety by banning nudifiers." Both descriptions are incomplete. The larger operational story is that the EU is trying to prevent a compliance cliff.

The Council said in March that the Commission wanted the delay because the needed standards and tools might not be ready by the original deadline.[3] Parliament's committee readout said the same thing more directly: key standards may not be finalised by 2 August 2026, so fixed later dates were needed for predictability and legal certainty.[4] The Commission's May 7 statement repeats that logic by arguing the new sequencing will help ensure standards and other support tools are in place before the rules start to apply.[1]

That means the policy signal is not "less AI law" in the abstract. It is more specific: Brussels is prioritising implementability. High-risk obligations are expensive only partly because the law exists. They are expensive because companies need harmonised standards, guidance, product-safety boundaries, and supervisory clarity to know whether a system is high-risk at all and how to document compliance once it is.[1][2][3][4][5]

The nudifier ban, meanwhile, shows what lawmakers did not want to postpone. The deal bans placing on the EU market AI systems intended to generate non-consensual intimate content, placing systems on the market without reasonable safety measures to prevent that use, and deploying such systems for that purpose.[2] In other words, the EU chose to slow the heavy compliance machinery while making one politically legible harm category more explicit.

Decision impact

Next 24 hours

Teams selling or deploying AI in the EU should treat the political deal as a strong planning signal, but not as the final legal text yet. Compliance calendars that assumed August 2026 for all high-risk cases now need a branch plan rather than a single countdown.[1][2]

Next 7 days

Product counsel, standards teams, and procurement leads should map which systems fall into the two delayed buckets:

stand-alone high-risk use cases such as biometrics, critical infrastructure, education, employment, law enforcement, migration, asylum, or border control
AI systems embedded in products governed by sectoral safety legislation[1][2][3][4]

This is also the right window to review whether any image, audio, or video generation workflows will face the 2 December 2026 watermarking date if the deal is formally adopted.[2]

Next 30 days

The live risk is procedural, not conceptual. Watch whether Parliament and Council move quickly enough to complete final adoption before 2 August 2026.[2] If the text stalls, companies cannot assume the old deadline has disappeared. If the text moves smoothly, the practical benefit is time: more room for standards work, guidance, and internal classification decisions.[1][2][5]

Scenario map

Base case: the provisional deal is formally adopted before August 2, 2026, and the market resets its main high-risk compliance plans to late 2027 and 2028. Trigger: Parliament and Council complete the legislative steps on the current timetable.[2]
Upside case: the delay is followed by clearer Commission guidance, more usable standards, and less fragmentation between product-safety law and AI Act obligations. Trigger: the post-adoption guidance and standards package arrives early enough to turn the extra time into real compliance clarity rather than drift.[1][3][5]
Downside case: formal adoption slips, companies slow their preparations too aggressively, and the old August 2026 risk window remains operationally relevant longer than expected. Trigger: legislative slippage or ambiguous transitional communication from institutions.[2][5]

Action checklist

Rebuild any EU AI compliance roadmap that still treats 2 August 2026 as a fixed final date for every high-risk system.
Separate stand-alone high-risk systems from product-embedded systems; the new proposal gives them different clocks.[1][2][3][4]
Review generative-media workflows for the likely 2 December 2026 watermarking date.[2]
Treat the nudifier ban as a near-term policy priority area even though the broader high-risk timeline is moving.[2][3][4]
Keep one explicit invalidation condition in internal plans: if final adoption slips, the old deadline may still matter.[2]

cronfeed.work