Your compliance calendar probably still has August 2, 2026 circled as the day full EU AI Act high-risk obligations hit. That date is no longer correct, and treating it as either still-urgent or now-irrelevant are both mistakes. On June 29, 2026, the Council of the EU gave final approval to the Digital Omnibus on AI, locking in a 16-month deferral for standalone high-risk AI systems and a 12-month deferral for AI embedded in regulated products. The European Parliament had already endorsed the package on June 16, 2026, by a 423-57 vote with 174 abstentions. Publication in the Official Journal is expected in July 2026, with the amended regulation entering into force three days after that.
Here is what that actually means: the deadline moved, but almost nothing else did.
Key Takeaways
- Standalone high-risk AI systems under Annex III (hiring tools, credit scoring, biometric identification, and similar) now have until December 2, 2027 to comply, not August 2, 2026: a 16-month extension confirmed by the Council on June 29, 2026.
- High-risk AI embedded in regulated products under Annex I (medical devices, machinery, toys with AI components) now have until August 2, 2028, a 12-month extension from the original August 2027 date.
- Prohibited AI practices have been enforceable since February 2, 2025, and general-purpose AI model obligations have applied since August 2, 2025. Neither date moved.
- A new prohibited practice was added to Article 5: AI systems that generate or manipulate non-consensual intimate imagery or CSAM ("nudifier" apps) are now banned outright, with a transitional watermarking rule running to December 2, 2026, for systems already on the market before August 2, 2026.
- Article 50 transparency obligations (telling users they're talking to a chatbot, labeling deepfakes) are unaffected and still apply from August 2, 2026.
- Until formal publication in the Official Journal, the original August 2026 date remains the technical legal baseline. Treat December 2027 as the very likely outcome, not yet a certainty to build a compliance pause around.
What Actually Happened, in Order
The Digital Omnibus on AI did not appear out of nowhere. The European Commission published the proposal on November 19, 2025, as part of a broader package aimed at simplifying GDPR, the AI Act, and the Product Liability Directive. The Council and Parliament reached a provisional political agreement on May 7, 2026, and Executive Vice-President Henna Virkkunen framed it as a dual mandate: "Our businesses and citizens want two things from AI rules. They want to be able to innovate and feel safe. Today's agreement does both."
From there, the process moved through the EU's ordinary legislative steps in rapid succession:
- May 7, 2026: Council and Parliament negotiators reach provisional political agreement.
- June 16, 2026: European Parliament formally adopts the text, 423 votes in favor, 57 against, 174 abstentions.
- June 29, 2026: Council of the EU gives final approval, completing the legislative process on the Council side.
- Expected July 2026: Publication in the Official Journal of the European Union, with entry into force three days later.
That is a five-month sprint from proposal to final approval, unusually fast for EU legislation, and a sign of how urgently both institutions wanted certainty locked in before the original August 2, 2026 deadline arrived. Co-rapporteur Arba Kokalari described the intent bluntly during the Parliament debate: "To all the entrepreneurs and engineers out there, we are pressing the pause button on the AI Act and we are reducing red tape." That framing is accurate for the compliance calendar. It is not accurate for the substance of what high-risk providers and deployers will eventually have to do: the requirements themselves did not shrink.
One important caveat belongs up front, not buried in an FAQ: formal legal effect depends on Official Journal publication. Multiple law firm trackers, including Gibson Dunn, note that these changes "only take legal effect upon formal adoption and publication of the Omnibus in the Official Journal." Council approval is the last political hurdle, not the last procedural one. Publication has not happened as of this writing. Treat the new dates below as the outcome you should plan around, while keeping an eye out for the Official Journal notice that makes it official.
The Two New Deadlines, by AI System Category
The Digital Omnibus splits high-risk AI systems into the same two categories the AI Act always used, Annex III and Annex I, but gives each one a different amount of extra runway.
| Annex III: standalone high-risk systems | Annex I: high-risk AI embedded in products | |
|---|---|---|
| Examples | Hiring tools, credit scoring, biometric identification, education access, migration and border control, law enforcement | Medical devices, machinery, toys, lifts with embedded AI |
| Original deadline | August 2, 2026 | August 2, 2027 |
| New deadline | December 2, 2027 | August 2, 2028 |
| Extension length | 16 months | 12 months |
| Compliance path | Article 9-15 obligations: risk management, data governance, technical documentation, human oversight, EU database registration | Conformity assessment under existing product-safety legislation, via notified bodies |
Annex III: standalone high-risk AI systems, now December 2, 2027. This is the larger category for most Secure Privacy readers: AI used in recruitment and employee management, creditworthiness assessment, biometric identification and categorization, education and vocational training access, migration and border control, law enforcement, and administration of justice. These systems were originally due to meet the full slate of Article 9-15 obligations by August 2, 2026. That date is now December 2, 2027, a 16-month extension.
Annex I: high-risk AI embedded in regulated products, now August 2, 2028. This category covers AI that is a safety component of, or is itself, a product already regulated under EU product-safety law. These systems follow the conformity-assessment procedures of their underlying product legislation, and their original deadline of August 2, 2027, has moved 12 months to August 2, 2028.
The gap between the two dates is deliberate. Annex III systems are usually pure software deployed directly by the organization using them, so the Commission and co-legislators judged that classification and governance work could reasonably compress into a shorter runway once the tooling and standards catch up. Annex I systems ride on top of product-safety conformity assessment regimes that involve notified bodies and longer certification cycles, which is why they retain the longer clock.
If your organization uses AI in hiring, credit decisions, or similar Annex III contexts, classifying which of your AI systems actually fall into a high-risk category is still the first task on the list. The extension changes when you must finish, not whether you need to start.
The stakes behind both deadlines have not softened. Breaches of high-risk system requirements, once the applicable date arrives, still carry fines of up to €15 million or 3% of total worldwide annual turnover, whichever is higher. The Digital Omnibus did add one piece of relief worth knowing about: it introduces formal definitions for small and medium-sized enterprises and small mid-cap companies inside the AI Act itself, and extends them simplified technical documentation, proportionate quality-management requirements, and priority access to regulatory sandboxes. That relief affects how much a smaller organization has to produce, not whether the December 2027 and August 2028 dates apply to it.
What Does Not Change
This is the part that gets lost in headlines about a "delay," and it's worth stating plainly because getting it wrong in either direction creates real exposure.
Prohibited AI practices have been enforceable since February 2, 2025, and stay that way. Social scoring, manipulative AI that exploits vulnerabilities, real-time remote biometric identification in public spaces (subject to narrow law-enforcement exceptions), and untargeted scraping of facial images to build recognition databases are all still banned today, with no change to that date anywhere in the Digital Omnibus package.
General-purpose AI model obligations have applied since August 2, 2025, and were left alone. If your organization develops, fine-tunes, or places a general-purpose AI model on the EU market, the transparency, technical documentation, and copyright-policy obligations under Articles 51-56 have already been running for nearly a year and were not touched by this legislative package. Downstream companies that merely integrate a GPAI model via API, without preserving or extending its general-purpose character, do not carry these obligations themselves, but that distinction predates the Omnibus and is unaffected by it.
Article 50 transparency obligations still apply from August 2, 2026. Providers of chatbots and generative AI systems still must disclose that users are interacting with AI. Providers of systems that generate synthetic audio, image, video, or text must still mark that output as artificially generated, and deepfake content still requires disclosure. None of that moved. The only change inside Article 50 is a narrow four-month grace period, described below, that applies exclusively to the machine-readable watermarking format for systems already on the market before the Act's transparency rules took effect.
The practical read, echoed across the law firm analyses tracking this package, is consistent: the deferral buys time to build the compliance infrastructure correctly, it does not relieve anyone of the underlying duty to build it. As Morgan Lewis put it in its client alert on the amendments, regulators "have not signaled tolerance for pause" on the substance, only more runway to implement it without rushing.
Not sure whether a given system in your stack even qualifies as high-risk in the first place? That classification question is worth settling before the extended clock, not the old one, becomes the thing you're racing.
The New Article 5 Prohibition: AI-Generated CSAM and Non-Consensual Intimate Imagery
The Digital Omnibus did not only push deadlines back. It added a new category to the Article 5 list of prohibited AI practices, and this is the part of the package least likely to be covered by generic "deadline delayed" headlines.
The new prohibition bans placing on the market, putting into service, or using an AI system that generates or manipulates non-consensual intimate images, video, audio, or similar material, including AI-generated child sexual abuse material (commonly discussed under the "nudifier app" label). The scope is written broadly: it captures not just systems built specifically for that purpose, but any system where generating such content is a reasonably foreseeable and reproducible outcome that does not require significant technical modification to achieve, and where the system lacks adequate technical safeguards against it.
Co-rapporteur Michael McNamara was direct about why the provision exists, telling the European Parliament that these systems "impact real people, overwhelmingly women, with the purpose of humiliating, degrading and objectifying them." Unlike the Annex III and Annex I deadline extensions, this is a new restriction, not a delayed one — a rare instance in this package of the rules getting stricter rather than looser.
There is one transitional element attached to it. Providers of AI systems that generate synthetic content and were already placed on the market before August 2, 2026, get until December 2, 2026 (a four-month grace period) to implement the machine-readable watermarking and detection format that Article 50(2) requires for that content. New systems entering the market after August 2, 2026, do not get this grace period; the watermarking requirement applies to them immediately.
Organizations using generative AI tools for marketing content, customer-facing chatbots, or synthetic media production should treat December 2, 2026, as a hard date for output-marking compliance, separate from and unaffected by the Annex III/Annex I extensions discussed above.
Why You Should Not Stop Preparing
It's tempting to read "16-month extension" as "16 months of breathing room to do nothing." That reading gets the arithmetic right and the strategy wrong.
The obligations that become mandatory on December 2, 2027, have not been simplified. A provider or deployer of a high-risk Annex III system will still need a functioning risk management system, data governance documentation, technical files, human oversight procedures, and a completed Fundamental Rights Impact Assessment where required: the same list that existed before the Omnibus, just with a later hard stop. Building that infrastructure retroactively, under deadline pressure, inside a system that has been running in production for a year or more, is materially harder than building it in from the start. Organizations that treat the extension as a reason to start now and iterate, rather than a reason to wait until late 2027, will spend less overall and produce fewer scramble-driven errors.
There is also a narrower, more immediate risk: organizations that already built AI governance programs around the original August 2, 2026 date, and that pause or dismantle that work because the deadline moved, will need to remobilize it later, often with less institutional memory of why decisions were made the first time. A governance maturity program that tracks AI governance alongside privacy and cybersecurity maturity gives you a documented baseline to pick back up from, rather than a program that quietly evaporates during the extension and has to be rebuilt from scratch closer to 2027.
Secure Privacy's AI Governance module is built for exactly this continuity problem: it maintains a registry of AI systems with risk levels, use cases, and responsible owners, maps each system to applicable regulations including the EU AI Act, and flags high-risk deployments automatically as your inventory changes, so classification work done today under the original timeline still counts under the revised one. The platform's Assessments module runs Fundamental Rights Impact Assessments, DPIAs, and related evaluations from a single interface, and its compliance calendar surfaces the revised statutory dates (December 2, 2027 for Annex III; August 2, 2028 for Annex I; December 2, 2026 for the watermarking transition) so a moved deadline doesn't silently fall off a compliance team's radar.
A Note on What You May Have Read Elsewhere
If you encountered coverage earlier in 2026, including in some of our own prior analysis, treating August 2, 2026 as the fixed, non-negotiable deadline for Annex III high-risk AI systems, that reflected the accurate legal status at the time it was written, before the Council's June 29, 2026 final approval. It is now outdated. The Digital Omnibus was, for most of the first half of 2026, a proposal moving through trilogue negotiations with an uncertain outcome and no fixed adoption date, which is exactly why earlier guidance was right to caution against treating an unconfirmed extension as settled. That uncertainty resolved on June 29, 2026. If you are working from anything published before that date, the underlying legal status of the extension has changed from "proposed" to "adopted, pending Official Journal publication," and your compliance calendar should be updated accordingly.
Frequently Asked Questions
Is the EU AI Act high-risk deadline extension final law yet?
Not fully. The Council gave final approval on June 29, 2026, following the European Parliament's endorsement on June 16, 2026, but the changes take formal legal effect only once the text is published in the Official Journal, expected in July 2026, and the amended regulation enters into force three days after that.
What is the new deadline for high-risk AI systems under the EU AI Act?
Standalone high-risk AI systems under Annex III now have until December 2, 2027, a 16-month extension from the original August 2, 2026 date. High-risk AI embedded in regulated products under Annex I now have until August 2, 2028, a 12-month extension from August 2, 2027.
Does the extension apply to the AI Act's prohibited practices or GPAI rules?
No. Prohibited AI practices have been enforceable since February 2, 2025, and general-purpose AI model obligations have applied since August 2, 2025. Neither date changed under the Digital Omnibus.
What is the new Article 5 prohibition the Digital Omnibus added?
A ban on AI systems that generate or manipulate non-consensual intimate images, video, or audio, including AI-generated child sexual abuse material. Systems already on the market before August 2, 2026, get a transitional period until December 2, 2026, to implement the required machine-readable watermarking; new systems must comply immediately.
Should organizations pause EU AI Act compliance work because of the extension?
No. The extension changes the compliance date, not the underlying obligations. Risk management, data governance, technical documentation, human oversight, and impact assessment requirements for Annex III systems are unchanged in substance. Only the enforcement date moved, to December 2, 2027.
Does the deadline extension affect Article 50 transparency obligations?
No. Chatbot disclosure and AI-generated content labeling obligations under Article 50 remain on their original schedule of August 2, 2026. The only related change is a four-month grace period, to December 2, 2026, for the machine-readable watermarking format, and only for systems already on the market before August 2026.
