TechAILatest.newsTrackers
AI Policy & Regulation / EU AI Act

EU AI Act after the Omnibus: what changes on 2 December 2026 for companies?

The AI Omnibus moved EU AI Act dates. What applies on 2 December 2026, which high-risk duties now wait until 2027 and 2028, and what companies should check.

By Emre A.Published

Abstract illustration of the columns of a public institutionTech AI Latest
Illustration of the columns of a public institution, made by Tech AI Latest for this story. Illustration: Tech AI Latest (Original illustration)

The AI Omnibus, Regulation (EU) 2026/1744, rewrote the EU AI Act timetable in July 2026. Two things land on 2 December 2026: a new ban on AI systems that generate non-consensual intimate images or child sexual abuse material, and the end of the grace period for marking AI output from generative systems already on the market. High-risk obligations moved to 2 December 2027 (Annex III) and 2 August 2028 (products under Annex I).

Key facts

  • Amending law: Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026, in force from 27 July 2026
  • 2 December 2026: new Article 5 ban on AI that generates non-consensual intimate content or child sexual abuse material
  • 2 December 2026: deadline for machine-readable marking (Article 50(2)) by generative systems placed on the market before 2 August 2026
  • High-risk AI under Annex III (hiring, credit, education and similar): from 2 December 2027 instead of 2 August 2026
  • High-risk AI built into regulated products (Annex I): from 2 August 2028 instead of 2 August 2027
  • Unchanged: chatbot disclosure and deepfake labelling under Article 50 have applied since 2 August 2026

Why the EU AI Act timetable changed

The EU AI Act, Regulation (EU) 2024/1689, entered into force on 1 August 2024 and was designed to apply in stages: bans on prohibited practices from February 2025, duties for general-purpose AI models from August 2025, and most of the remaining rules, including those for high-risk systems, from 2 August 2026. By early 2026 it was clear that the harmonised technical standards companies need to prove compliance would not be ready in time. The first one, EN 18286 on quality management systems, was only cited in the Official Journal on 14 October 2026.

The European Commission's answer was the Digital Omnibus on AI, part of a wider package meant to simplify EU digital rules. After negotiations between the Parliament and the Council, the text was adopted as Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026 and in force from 27 July 2026, as the Commission confirmed when it entered into force. It does not replace the AI Act; it amends specific articles and dates. The amending regulation is the legal text; law firms including Gibson Dunn, White and Case and Cooley, and the European Parliament's research service, have published summaries of it, and this explainer relies on points where those summaries agree.

The result is a law that is partly in force, partly postponed and partly extended with new rules. That makes the calendar the first thing a compliance team needs to get right, because the dates that most companies planned around in 2025 are no longer the dates that apply.

What applies on 2 December 2026

Two obligations start on 2 December 2026, eighteen days from the publication of this explainer.

The first is a new prohibition added to Article 5, the list of banned AI practices. It covers AI systems that generate non-consensual intimate or sexually explicit images, video or audio of real people, and systems that generate child sexual abuse material. Commentators describe it as aimed at so-called nudification apps, tools whose purpose is to undress people in photos. Unlike most of the AI Act, there is no grace period for systems that are already on the market. Prohibited practices carry the highest fines in the regulation, up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher, under Article 99. Deepfake abuse has been a recurring theme in enforcement elsewhere: Interpol ran a campaign against deepfake scams this autumn, and the FBI warned about AI voice impersonation before the US midterms.

The second is the end of a transition period for Article 50(2). That provision requires providers of AI systems that generate synthetic audio, images, video or text to mark the output in a machine-readable format so it can be detected as AI-generated. Article 50 as a whole applies from 2 August 2026, but the Omnibus gave generative systems placed on the market before that date until 2 December 2026 to add the marking. Systems launched on or after 2 August 2026 have had to comply from launch. In practice this means that image generators, video models, voice cloning tools and text generators that were already offered in the EU in July 2026 need working watermarks, metadata or equivalent techniques by the start of December.

To help with that, the Commission published a voluntary Code of Practice on marking and labelling AI-generated content on 10 June 2026. It has two parts: one for providers, on marking and detecting synthetic content, and one for deployers, on labelling deepfakes and AI-generated text published on matters of public interest. Its annex introduces common EU icons for labels. Signing the code is not mandatory, but law firms expect regulators to treat it as the reference for what compliance looks like.

What moved to 2027 and 2028

The biggest change is the delay of the high-risk rules. AI systems listed in Annex III, the stand-alone uses the law treats as high-risk, such as recruitment and worker management, creditworthiness checks, access to education, essential public services and some law enforcement uses, now face their obligations from 2 December 2027 instead of 2 August 2026. AI built into products that are already regulated by EU safety law, listed in Annex I, such as machinery, medical devices and toys, now has until 2 August 2028 instead of 2 August 2027.

The Commission's original proposal tied the start of the high-risk rules to the availability of standards and support tools, which would have left the date uncertain. The final text uses fixed calendar dates instead, according to the summaries of the adopted regulation. For companies that is easier to plan around: a provider of a hiring tool now knows it has a little over a year, not an open-ended wait.

Some dates did not move. The duties for general-purpose AI models, which cover the large language models from developers such as OpenAI, Google and Anthropic, have applied since 2 August 2025, and the Commission's power to fine model providers has been available since 2 August 2026. The transparency duties in Article 50 other than the marking grace period, telling people they are talking to a chatbot and labelling deepfakes, have also applied since 2 August 2026.

What else the Omnibus changed

Article 4 on AI literacy was softened. The original text required providers and deployers to ensure a sufficient level of AI literacy among their staff. The amended text asks them to take measures that support AI literacy, and gives the Commission and member states a role in promoting it, particularly for small companies. It does not set a required literacy level for any individual. The obligation still exists, so a training programme remains the simplest way to show compliance.

Relief that the AI Act already gave small and medium-sized enterprises, such as simplified technical documentation, now extends to small mid-cap companies. These are defined by Commission Recommendation (EU) 2025/1099 as companies with fewer than 750 employees and either turnover of up to EUR 150 million or a balance sheet of up to EUR 129 million.

The rules on scope did not change. The AI Act still applies to companies outside the EU whose systems or outputs are used in the EU, and it now applies in a limited form in Northern Ireland from 10 November 2026, but only to AI built into physical products.

What companies should check before December

Our reading of the amended calendar gives a short list. Any company that offers image, video, audio or text generation in the EU should confirm that its output carries machine-readable marking by 2 December 2026, and that it can show which technique it uses; the Code of Practice is the practical template. Any company that runs or hosts generative tools, including app stores and platforms that distribute third-party apps, should check that nothing it offers falls under the new ban on non-consensual intimate content, because there is no transition period and the fines are the highest in the law.

Companies building or buying high-risk systems gained time but not certainty about what auditors will expect. The first harmonised standard exists; more are being drafted by CEN and Cenelec. The months before December 2027 are the window to build the risk management, documentation, data governance and human oversight the law requires, and to watch the EU AI Scientific Panel and the AI Office for guidance. Businesses that automate hiring or credit decisions with tools from vendors such as SAP should ask those vendors now which obligations the vendor will carry as provider and which fall on them as deployer.

For a broader view of the models these rules apply to, our guides on AI model API pricing and open-weight AI models list current releases, and all guides are on the guides page.

Questions people ask

What happens under the EU AI Act on 2 December 2026?
A new ban on AI systems that generate non-consensual intimate content or child sexual abuse material applies, and generative AI systems placed on the EU market before 2 August 2026 must mark their output in a machine-readable way under Article 50(2).
When do the EU AI Act high-risk rules apply now?
From 2 December 2027 for stand-alone high-risk systems in Annex III, such as hiring and credit scoring, and from 2 August 2028 for AI built into regulated products in Annex I, under Regulation (EU) 2026/1744.
Was the AI Act watermarking rule delayed?
Only for systems already on the market before 2 August 2026, which have until 2 December 2026. New systems have had to mark output since 2 August 2026, and chatbot disclosure and deepfake labelling were not delayed.
Does the AI Omnibus remove the AI literacy duty?
No. It softens Article 4: providers and deployers must take measures to support staff AI literacy rather than ensure a set level, and the Commission and member states must help promote it.

Sources

  1. AI Omnibus enters into force (European Commission, 27 Jul 2026)
  2. Regulation (EU) 2026/1744 amending the AI Act (AI Omnibus) (EUR-Lex, 24 Jul 2026)
  3. Commission publishes Code of Practice on marking and labelling AI-generated content (European Commission, 10 Jun 2026)
  4. Regulation (EU) 2024/1689 (AI Act) (EUR-Lex, 12 Jul 2024)

Reference pages

Spotted an error? See the corrections policy and tell the editor. Facts in this story come from the sources listed above; vendor claims are attributed.

Comments

Loading comments.

No links, no insults; comments that break the rules are not published. 0/1000