EU AI Act in 2026: What Applies Now, What Comes Next and a Practical Checklist

ApexlyTech Team 12 min read
EU AI Act in 2026: What Applies Now, What Comes Next and a Practical Checklist

TL;DR: The EU AI Act (Regulation (EU) 2024/1689) has applied in stages since 1 August 2024. Prohibited practices and AI literacy have applied since 2 February 2025, rules for general-purpose AI (GPAI) models since 2 August 2025, and most of the rest, including the Article 50 transparency duties and enforcement, since 2 August 2026. The big change this year is the Digital Omnibus on AI, adopted as Regulation (EU) 2026/1744 and in force since 27 July 2026. It moved the high-risk obligations to fixed later dates: 2 December 2027 for the Annex III use cases and 2 August 2028 for AI in regulated products. It also added two new prohibitions that apply from 2 December 2026. If you build or use AI in the EU, the work for this quarter is an inventory of your AI systems, a role and risk classification for each, transparency notices where people talk to or see output from AI, and a plan for any high-risk system that has to be ready by December 2027.

Where things stand on 29 October 2026

Much of what was written about the AI Act in 2025 now has the wrong dates. The table below reflects the Act as amended by the Digital Omnibus, as summarised on the Commission's own AI Act Service Desk timeline.

DateWhat appliesStatus
1 Aug 2024The Act enters into forceDone
2 Feb 2025Definitions, AI literacy (Article 4) and prohibited practices (Article 5)Applies
2 Aug 2025Obligations for providers of GPAI models; national authorities and penalty rules; EU governance bodiesApplies
2 Aug 2026Most remaining rules, including Article 50 transparency; enforcement starts at national and EU levelApplies
2 Dec 2026New prohibitions on AI that generates non-consensual sexual deepfakes or child sexual abuse material; end of the Article 50(2) grace period for generative systems already on the market before 2 Aug 2026Coming
2 Aug 2027GPAI models placed on the market before 2 Aug 2025 must comply; each Member State must have at least one AI regulatory sandbox runningComing
2 Dec 2027High-risk rules for the Annex III use casesComing (was 2 Aug 2026)
2 Aug 2028High-risk rules for AI in products covered by Annex I (Article 6(1))Coming (was 2 Aug 2027)

What the Digital Omnibus changed, and what it did not

The Commission proposed the Digital Omnibus on AI on 19 November 2025 as part of a wider simplification package. Its main motive was timing: the high-risk obligations were due to start before the harmonised standards and guidance that companies need to meet them were ready. The legislative path was quick by EU standards:

  • political agreement between Parliament and Council on 7 May 2026;
  • Parliament's position on 16 June 2026 and the Council's decision on 29 June 2026;
  • publication in the Official Journal on 24 July 2026 as Regulation (EU) 2026/1744, in force from 27 July 2026.

The proposal had linked the start of the high-risk obligations to the availability of standards, with long-stop dates. The adopted text replaced that mechanism with fixed dates: 2 December 2027 for Annex III and 2 August 2028 for Annex I. That is easier to plan around, and it means there is no further automatic slippage to wait for.

According to the Parliament's summary of the adopted act, the Omnibus also:

  • reframes the AI literacy duty, so providers and deployers support literacy efforts rather than having to guarantee a level of AI literacy for every individual;
  • extends the simplifications available to small and medium-sized enterprises (SMEs) to small mid-caps (SMCs);
  • narrows the definition of a safety component;
  • allows processing of special categories of personal data for bias detection and correction, under strict conditions;
  • widens the AI Office's exclusive supervision and enforcement powers;
  • adds the two new prohibitions listed above, from 2 December 2026.

What it did not change matters just as much. The prohibitions, the GPAI obligations and the Article 50 transparency duties were not postponed. The only transparency grace period is narrow: generative AI systems already on the market before 2 August 2026 have until 2 December 2026 to meet the Article 50(2) duty to mark their outputs in a machine-readable way.

First, work out your role

Almost every obligation in the Act attaches to a role, so classification starts here. The Act defines several, and one company can hold more than one for different systems:

  • Provider: develops an AI system or GPAI model, or has one developed, and places it on the market or puts it into service under its own name or trademark. A software company selling an AI feature is usually a provider.
  • Deployer: uses an AI system under its authority in a professional capacity. A business using an AI recruitment tool or chatbot is a deployer.
  • Importer and distributor: bring third-country AI systems onto the EU market, or make them available further down the chain.
  • Provider of a GPAI model: a separate category with its own chapter. Most companies that build applications on someone else's model are not GPAI model providers, but they may still be providers of the AI system they build.

Watch for role changes. If you put your name on a high-risk system, make a substantial modification to it, or change its intended purpose so that it becomes high-risk, the Act can treat you as its provider (Article 25).

The Act also reaches beyond the EU. It applies to providers outside the EU who place systems on the EU market, and to providers and deployers outside the EU where the output of the system is used in the EU (Article 2).

What already applies

Prohibited practices (Article 5)

Since 2 February 2025, a short list of uses has been banned outright. They include: manipulative or deceptive techniques that materially distort behaviour and cause significant harm; exploiting vulnerabilities due to age, disability or social or economic situation; social scoring; predicting that a person will commit a crime based solely on profiling or personality traits; building facial recognition databases through untargeted scraping of images from the internet or CCTV; emotion recognition in workplaces and education institutions (except for medical or safety reasons); biometric categorisation to infer sensitive characteristics such as race, political opinions or sexual orientation; and, with narrow exceptions, real-time remote biometric identification in publicly accessible spaces for law enforcement. From 2 December 2026 the list grows to include AI that generates non-consensual sexual deepfakes or child sexual abuse material. The Commission has published guidelines on the prohibited practices and on the definition of an AI system.

AI literacy (Article 4)

Providers and deployers have had to take measures on AI literacy since February 2025. The Omnibus softened this from an obligation of result to one of effort, but it did not remove it. In practice: know who in your organisation uses or builds AI, give them training proportionate to the risk, and keep a record of it.

General-purpose AI models

Since 2 August 2025, providers of GPAI models must keep technical documentation, give downstream providers the information they need, put in place a policy to comply with EU copyright law, and publish a summary of the content used for training using the Commission's template (Article 53). Models with systemic risk, presumed when training compute exceeds 1025 floating-point operations (Article 51), carry extra duties: model evaluations including adversarial testing, risk assessment and mitigation, serious incident reporting and cybersecurity (Article 55). The GPAI Code of Practice, published on 10 July 2025, is the route most large model providers have chosen to show compliance. The AI Office's enforcement powers over GPAI providers apply from 2 August 2026, and models placed on the market before 2 August 2025 have until 2 August 2027.

Transparency (Article 50)

From 2 August 2026, four transparency duties apply regardless of risk class:

  1. AI that interacts with people must be designed so that people are informed they are dealing with an AI system, unless that is obvious from the context (Article 50(1), a provider duty). This covers chatbots, voice agents and AI assistants inside your product.
  2. Generative AI producing synthetic audio, image, video or text must mark its outputs in a machine-readable format and make them detectable as artificially generated (Article 50(2), a provider duty).
  3. Emotion recognition and biometric categorisation: deployers must inform the people exposed to them (Article 50(3)).
  4. Deepfakes must be disclosed as artificially generated by the deployer, and AI-generated text published to inform the public on matters of public interest must be disclosed unless it has gone through human review and someone holds editorial responsibility (Article 50(4)).

The Commission has published guidelines on Article 50 and a voluntary Code of Practice on marking and labelling AI-generated content, finalised in June 2026, with separate sections for providers and deployers.

High-risk AI: what you now have until December 2027 to build

An AI system is high-risk in one of two ways. Either it is a safety component of, or itself, a product covered by the EU harmonisation legislation in Annex I (machinery, medical devices, toys, lifts and others) and needs third-party conformity assessment; or it falls into one of the Annex III use cases. Annex III covers biometrics; critical infrastructure; education and vocational training; employment and worker management (for example CV screening and promotion decisions); access to essential private and public services (for example creditworthiness assessment and life and health insurance pricing); law enforcement; migration, asylum and border control; and the administration of justice and democratic processes.

Article 6(3) lets a provider conclude that an Annex III system is not high-risk when it does not pose a significant risk of harm, for example because it performs a narrow procedural task or only improves the result of a completed human activity. A system that profiles natural persons is always high-risk. If you rely on this exception, document the assessment before placing the system on the market. The Commission's guidelines on high-risk classification, which give practical examples, were due in early 2026 and are worth checking for the latest version before you finalise any assessment.

For providers, the high-risk requirements are substantial and mostly engineering work:

  • a risk management system across the life cycle (Article 9);
  • data governance for training, validation and test data, including examination for bias (Article 10);
  • technical documentation (Article 11) and automatic logging of events (Article 12);
  • instructions for use that tell deployers what the system can and cannot do (Article 13);
  • design for effective human oversight (Article 14);
  • appropriate accuracy, robustness and cybersecurity (Article 15);
  • a quality management system (Article 17), conformity assessment, CE marking and registration in the EU database;
  • post-market monitoring and reporting of serious incidents (Articles 72 and 73).

Deployers of high-risk systems have their own duties under Article 26: use the system according to its instructions, assign human oversight to people with the competence and authority to exercise it, monitor operation, keep the logs under their control, and inform workers before using high-risk AI in the workplace. Public bodies and some private deployers, including those assessing creditworthiness or pricing life and health insurance, must also carry out a fundamental rights impact assessment (Article 27).

On standards, the first harmonised standard supporting the Act, EN 18286 on quality management systems, was referenced in the Official Journal by Commission Implementing Decision (EU) 2026/2274 in October 2026. Standards on risk management, data governance, logging, accuracy, robustness and cybersecurity are still being developed by CEN and CENELEC. Using a harmonised standard gives a presumption of conformity; until the rest arrive, build to the text of the articles and keep your design decisions documented.

Penalties and enforcement

Article 99 sets maximum fines of EUR 35 million or 7 percent of worldwide annual turnover (whichever is higher) for prohibited practices; EUR 15 million or 3 percent for most other obligations; and EUR 7.5 million or 1 percent for supplying incorrect or misleading information to authorities. For SMEs, including start-ups, the lower of the two figures applies. GPAI model providers face fines of up to EUR 15 million or 3 percent from the Commission (Article 101). National market surveillance authorities enforce most of the Act; the AI Office supervises GPAI models, and the Omnibus widened its exclusive powers further.

A practical checklist

  1. Build an AI inventory. List every AI system you develop, sell or use, including AI features inside third-party SaaS. For each, record the purpose, the model and supplier behind it, the data it uses, who is affected and which EU markets it reaches.
  2. Assign a role per system. Provider, deployer, importer, distributor or GPAI model provider. Note where your own changes could make you the provider.
  3. Screen against Article 5. Confirm nothing in the inventory is prohibited, including the two new categories from 2 December 2026. Pay particular attention to emotion recognition at work and biometric features.
  4. Classify risk. Check every system against Annex I and Annex III. Document any Article 6(3) "not high-risk" conclusion with your reasoning.
  5. Ship transparency now. Add clear AI disclosures to chatbots, voice agents and assistants; make sure generative features mark their outputs; label deepfakes and AI-written public-interest text as Article 50 requires.
  6. Document AI literacy. Short, role-based training for the people who build, buy, configure and supervise AI, with a record of who completed it.
  7. Review supplier terms. If you build on a third-party model, get the documentation the GPAI chapter obliges its provider to give you, and check the contract covers data use, retention and incident notification.
  8. Plan high-risk work backwards from 2 December 2027. Risk management, data governance, logging, human oversight design and a quality management system take months. Start with a gap assessment against Articles 9 to 17, and track the harmonised standards as they are published.
  9. Set up incident and change processes. Decide who assesses a serious incident, how logs are retained and how a model change is reviewed before release.
  10. Assign an owner and review quarterly. Guidance is still arriving. Someone should own the inventory and re-check it against new Commission guidance and standards each quarter.

How ApexlyTech can help

We build AI features and automation for clients and run them on our own platforms, including Apex AI. We can help you put the engineering side of this checklist in place: an AI inventory, disclosure and output marking in your product, logging and human-oversight controls, and documentation that keeps pace with your releases. See our AI and automation and custom software services, or contact us. For legal interpretation of your specific case, work with qualified counsel; this guide is not legal advice.

Frequently asked questions

Was the EU AI Act delayed?

Partly. The Digital Omnibus on AI, Regulation (EU) 2026/1744, moved the high-risk obligations to 2 December 2027 for Annex III systems and 2 August 2028 for AI in Annex I products. The prohibitions, GPAI obligations and Article 50 transparency duties were not delayed.

Does the AI Act apply to companies outside the EU?

Yes, where they place AI systems or GPAI models on the EU market, or where the output of their AI system is used in the EU. Providers established outside the EU may need an authorised representative in the EU for certain obligations.

Do I need to tell customers they are talking to a chatbot?

Yes. Since 2 August 2026, Article 50(1) requires AI systems that interact directly with people to be designed so that people are informed they are dealing with AI, unless it is obvious from the context.

We only use ChatGPT-style tools internally. Does anything apply?

Likely the AI literacy duty and the prohibitions, and the transparency rules if you publish AI-generated content or deploy it to talk to customers. Internal use in an Annex III area, such as screening job applicants, can make you a deployer of a high-risk system from December 2027.

What are the maximum fines?

Up to EUR 35 million or 7 percent of worldwide annual turnover for prohibited practices, EUR 15 million or 3 percent for most other breaches, and EUR 7.5 million or 1 percent for misleading information to authorities. SMEs pay the lower of the two figures.

Sources