Compliance & funding
EU AI Act Timeline 2026–2028 After the AI Omnibus
EU AI Act timeline 2026–2028 after the AI Omnibus: every deadline in one table, risk classes for common business AI uses and a 10-step SME compliance roadmap.

Key takeaways
In short:
- The AI Omnibus (Regulation (EU) 2026/1744) has been in force since 27 July 2026. It moves the high-risk obligations to 2 December 2027 (Annex III, e.g. recruitment, credit scoring) and 2 August 2028 (Annex I, AI in regulated products).
- Not postponed: prohibitions and AI literacy have applied since 2 February 2025, the rules for general-purpose AI (GPAI) models since 2 August 2025, and the Article 50 transparency rules (chatbots, deepfakes) since 2 August 2026.
- On 2 December 2026 a new prohibition (AI-generated child sexual abuse material and non-consensual intimate content) and the marking deadline for generative AI already on the market take effect.
- Enforcement is national: every EU country designates its own market surveillance authorities, which is why practice will differ across Europe.
- For most SMEs, the work right now is an AI inventory, Article 4 training and disclosures for chatbots and AI-generated content. HR and scoring tools need a plan before the end of 2027.
The EU AI Act (Regulation (EU) 2024/1689) is the EU's first comprehensive law on artificial intelligence. It applies in stages, and in summer 2026, just before the big 2 August milestone, the so-called Digital Omnibus on AI rearranged those stages. Many overviews online still show the old schedule. This guide reflects the position as of November 2026: which deadline applies when, what the Omnibus changed and what it left alone, how common business uses of AI are classified, and the order in which to tackle compliance.
This article is not legal advice. It does not replace a review of your specific use case by a qualified lawyer.
What are the EU AI Act deadlines for 2026, 2027 and 2028?
The table below brings together every milestone relevant to businesses as amended by the AI Omnibus. The legal basis is Article 113 (application dates) and Article 111 (transitional rules) in the consolidated text, together with the European Commission's AI Act Service Desk timeline.
| Date | What applies from this day | Before the Omnibus | Who is affected |
|---|---|---|---|
| 1 Aug 2024 | AI Act enters into force, no obligations yet | unchanged | – |
| 2 Feb 2025 | Definitions, AI literacy (Art. 4), prohibitions (Art. 5) for eight practices | unchanged | all providers and deployers |
| 2 Aug 2025 | Obligations for GPAI models (Chapter V), governance, penalty rules (Art. 99), notified bodies | unchanged | model providers, Member States |
| 27 Jul 2026 | AI Omnibus in force: new Art. 4 wording, new Art. 4a (bias detection), relief for small mid-caps | – | everyone |
| 2 Aug 2026 | General application: Art. 50 transparency rules, enforcement by national authorities and the AI Office | unchanged (Annex III high-risk was originally due here) | providers and deployers of chatbots, voicebots, generative AI |
| 2 Dec 2026 | New prohibition on AI generating CSAM and non-consensual intimate content; Art. 50(2) for generative systems placed on the market before 2 Aug 2026 | new | providers of generative AI |
| 1 Aug 2027 | Commission guidelines on high-risk classification (Art. 6) and on Art. 8(2), 9(10), 17(3) due | new | providers of high-risk AI |
| 2 Aug 2027 | GPAI models placed on the market before 2 Aug 2025 must comply; at least one AI regulatory sandbox per Member State | unchanged | model providers, authorities |
| 2 Sep 2027 | Commission guidance on post-market monitoring plans | new | providers of high-risk AI |
| 2 Dec 2027 | High-risk obligations for Annex III (incl. employment, education, creditworthiness, biometrics) | 2 Aug 2026 | providers and deployers, e.g. of HR and scoring tools |
| 2 Aug 2028 | High-risk obligations for Annex I (AI in products such as machinery, medical devices, toys) | 2 Aug 2027 | manufacturers of regulated products |
| 2 Aug 2030 | Existing high-risk systems used by public authorities must comply | unchanged | public bodies and their suppliers |
The dates for Commission guidance come from Article 96, the Future of Privacy Forum analysis and the artificialintelligenceact.eu timeline. The legislative steps of the Omnibus are documented in the European Parliament's Legislative Observatory.
Important: the grandfathering rule for existing high-risk systems
Under Article 111(2), high-risk rules apply to systems placed on the market or put into service before the relevant date only if they are significantly changed in their design afterwards. Systems used by public authorities must comply by 2 August 2030 at the latest. Do not rely on this blindly: many SaaS tools are updated continuously, and whether an update is "significant" is usually assessed by the vendor, not by you.
What is the AI Omnibus and how was it adopted?
The AI Omnibus (officially the "Digital Omnibus on AI") is an amending regulation to the AI Act. The Commission tabled proposal COM(2025) 836 on 19 November 2025. Parliament and Council reached a political agreement on 7 May 2026. Parliament voted on 16 June 2026 and the Council adopted the text on 29 June 2026. It was signed on 8 July and published in the Official Journal on 24 July 2026 as Regulation (EU) 2026/1744 (procedure: Legislative Observatory). According to the European Commission it entered into force on 27 July 2026, six days before the original milestone.
The reason for the delay: harmonised standards and guidance for high-risk AI were not ready in time. The Commission says the new dates ensure the rules apply once businesses have standards and other support tools. The guidelines on high-risk classification were still only a draft in May 2026, with consultation open until 23 July 2026.
The substantive changes at a glance
- New high-risk dates: 2 December 2027 for Annex III, 2 August 2028 for Annex I (Article 113).
- New prohibition from 2 December 2026: AI systems that generate child sexual abuse material or non-consensual intimate content, such as "nudification" apps (European Commission). Article 5 now lists nine prohibited practices.
- Transition for Article 50(2): providers of generative AI whose systems were on the market before 2 August 2026 have until 2 December 2026 to add machine-readable marking (Article 111(4)).
- AI literacy softened: instead of "ensure", Article 4 now requires measures "to support the development of AI literacy" and states that no specific level has to be guaranteed for any individual. The Commission and Member States must support companies, particularly SMEs.
- Small mid-caps (SMCs): relief previously reserved for SMEs, such as simplified technical documentation and fines capped at the lower of the two amounts, now also covers companies with fewer than 750 employees and up to €150 million turnover (definition in Recommendation (EU) 2025/1099, per FPF). The Commission estimates that around 8,250 additional companies benefit.
- Bias detection: a new Article 4a allows providers and deployers of non-high-risk AI too to process special categories of personal data, under strict safeguards, to detect and correct bias (FPF).
- A stronger AI Office: it now supervises AI systems built on GPAI models from the same provider group and AI embedded in very large online platforms, and it may set up an EU-level regulatory sandbox (Article 57).
What did the Omnibus not change?
Many summaries skip these points, yet they matter most for SMEs:
- The risk categories and the Annex III list are unchanged. A recruitment screening tool is still high-risk; it just has more time.
- The eight original prohibitions have applied since 2 February 2025, including emotion recognition in workplaces and education (Article 5(1)(f)).
- Article 50 has applied as planned since 2 August 2026. Only the technical marking for systems already on the market (paragraph 2) got extra time.
- The AI literacy duty remains; only its standard is milder.
- The fine ceilings remain: up to €35 million or 7% of worldwide annual turnover for prohibited practices; up to €15 million or 3% for, among others, breaches of deployer obligations (Art. 26) and transparency obligations (Art. 50); up to €7.5 million or 1% for misleading information to authorities. SMEs face whichever amount is lower, and SMCs do too for the last two tiers (Article 99).
- The GDPR applies alongside the AI Act, unchanged. See the section on the data Omnibus below.
Who enforces the AI Act across Europe?
Enforcement is split. The AI Office at the European Commission enforces the rules for GPAI models. Everything else is handled by national market surveillance authorities designated by each Member State. Since 2 August 2026, both levels can enforce the applicable rules. Member States also set their own penalty rules within the ceilings of Article 99.
Designation has progressed unevenly. In July 2026 the Future of Privacy Forum reported that only a limited number of countries had notified their single points of contact to the Commission. In practice, this means:
- Which authority you deal with depends on the country and the sector. Financial supervisors, product safety authorities and data protection authorities may each have a role.
- Check the national contact point for every country in which you put AI into service. The Commission's AI Act Service Desk maintains a page of national resources.
- Example Germany: the implementing act was published in the Federal Law Gazette (BGBl. 2026 I No. 223) on 28 July 2026 and entered into force the following day. It makes the Bundesnetzagentur the central market surveillance authority and complaints office, while BaFin covers AI directly linked to regulated financial activities.
Which risk class applies to common business AI uses?
The AI Act has four tiers: prohibited, high-risk, transparency obligations (often called limited risk) and minimal risk, plus separate rules for GPAI models. What matters is the intended purpose, not the technology. Your role matters too: anyone who puts an AI system into service under their own name is a provider; anyone who simply uses it is a deployer (Article 3). The Article 4 AI literacy duty applies to every row in the table.
| Use case | Typical risk class | Main duties | Deadline |
|---|---|---|---|
| Website chatbot (FAQs, booking requests) | Transparency | Tell users they are talking to AI (Art. 50(1)); for bought-in products this is the vendor's duty, so secure it in the contract | since 2 Aug 2026 |
| AI phone assistant / voicebot | Transparency | AI disclosure at the first interaction at the latest, e.g. in the greeting (Art. 50(1) and (5)); synthetic voice marked machine-readably (Art. 50(2), provider) | since 2 Aug 2026; marking for legacy systems by 2 Dec 2026 |
| Lead scoring in the CRM (B2B) | usually minimal | no specific AI Act duties; GDPR (profiling, transparency) applies | – |
| CV screening, candidate ranking, targeted job ads by AI | high-risk (Annex III 4(a)) | deployer duties under Art. 26: use per instructions, human oversight, logs kept at least 6 months, inform workers and their representatives | 2 Dec 2027 |
| Creditworthiness assessment of individuals | high-risk (Annex III 5(b)) | Art. 26 plus a fundamental rights impact assessment (Art. 27); financial supervisors may be competent | 2 Dec 2027 |
| AI image generation for marketing | Transparency | provider marks outputs machine-readably (Art. 50(2)); you label deepfakes visibly (Art. 50(4)) | since 2 Aug 2026; legacy systems 2 Dec 2026 |
| Internal knowledge assistant (documents, RAG) | usually minimal | Art. 4, GDPR, access rights; AI nature is usually obvious internally | – |
| AI agents (emails, orders, bookings) | depends on the task | customer contact: Art. 50(1); decisions on applicants or credit: high-risk; repurposing an agent for a high-risk use can make you the provider (Art. 25) | 2 Aug 2026 or 2 Dec 2027 |
| Emotion analysis of employees (e.g. in a call centre) | prohibited (Art. 5(1)(f)) | do not use; exception only for medical or safety reasons | since 2 Feb 2025 |
Sources for the table: Annex III, Article 26, Article 27, Article 50 and Article 25. The classification is typical, not definitive; a specific feature set can lead to a different result.
Three classification traps
1. The provider trap. If you build a chatbot on top of a language model API and run it on your website under your own brand, you are the provider of that AI system and responsible for the Article 50(1) disclosure yourself. For a bought-in chatbot product, the duty sits with the vendor. As the deployer, you should still check that the disclosure actually appears.
2. The Article 6(3) exemption has limits. An Annex III system is not high-risk if it only performs a narrow procedural task or merely prepares a human assessment. But if it performs profiling of natural persons, Article 6 says it is always high-risk. A tool that filters applications by degree may therefore be assessed differently from one that predicts "culture fit".
3. Changing the purpose makes you the provider. If you deploy a general-purpose AI system, such as an agent built on a GPAI model, in a way that turns it into a high-risk system, Article 25(1)(c) treats you as the provider, with all provider obligations from conformity assessment to registration.
What should SMEs do now? A 10-step roadmap
The order follows the deadlines: what already applies first, then what is coming in 2027.
- Build an AI inventory. List every AI system, including AI features embedded in CRM, helpdesk, office and recruitment software. For each, record the purpose, vendor, users and affected people.
- Clarify your role. Are you a provider (built in-house, own brand, purpose changed) or a deployer? This decides the scope of your duties more than the technology does.
- Rule out prohibited practices. Check especially for emotion recognition of employees, manipulative design and biometric categorisation. These bans have applied since February 2025.
- Assign and document a risk class for each system, for example using the table above and the Commission's AI Act Service Desk resources. Write down your reasoning for borderline cases.
- Build AI literacy. Role-based training, a short AI policy and attendance records are a sound baseline. Our guide to the AI literacy obligation and AI policy shows what this looks like in practice.
- Implement transparency. A disclosure in the chatbot's first message, an announcement in the voicebot, visible labels on deepfakes. Wording and details are in our guide to AI Act Article 50 labelling. The Commission's Article 50 guidelines and the Code of Practice on AI-generated content explain the duties with examples.
- Question your vendors. Ask for their AI Act classification (including any documented Article 6(3) assessment), instructions for use, their plan for marking by 2 December 2026 and for high-risk conformity by 2 December 2027. Put it in the contract.
- Prepare high-risk deployments. For HR or scoring tools: assign human oversight to adequately trained people, keep logs for at least six months, inform workers and their representatives before use (Article 26(7)) and plan a fundamental rights impact assessment for credit scoring.
- Align with the GDPR. Review the data protection impact assessment, legal basis, transparency notices and rules on automated decisions in the same pass. Article 26(9) explicitly links the two.
- Assign ownership and a deadline calendar. Name a responsible person, enter the dates from the table and review new guidance every six months, such as the high-risk guidelines due by 1 August 2027.
Quick check: are you ready for 2 December 2026?
- Every chatbot and voicebot discloses AI at the first interaction at the latest.
- Your vendors for image, video, audio and text generation have confirmed how machine-readable marking is implemented.
- Deepfakes in ads and on social media are visibly labelled.
- None of your tools can generate intimate content of real people without consent.
- Article 4 training is documented, and new hires are trained during onboarding.
Digital Omnibus on the GDPR: where does it stand?
Alongside the AI Omnibus, the Commission presented a second, separate proposal on 19 November 2025: the Digital Omnibus COM(2025) 837. According to the European Parliament's procedure file, it would amend the GDPR, the ePrivacy Directive, the Data Act and the NIS2 Directive, among others. Public debate centres on AI training based on legitimate interest and on cookie rules.
As of November 2026, this proposal has not been adopted. The lead committees ITRE and LIBE presented a draft report in June 2026, and the status is still "Awaiting committee decision". For planning purposes, assume the current GDPR continues to apply, and treat reports of "relaxed data protection rules for AI" as proposals, not law.
Conclusion: 2027 is no reason to wait
The AI Omnibus has eased the pressure on high-risk AI, not on everything else. For most companies the key duties apply today: build and evidence AI literacy, rule out prohibited practices, and make chatbots, voicebots and AI-generated content transparent. If you use HR or scoring tools, do not wait for 2027. Choose vendors and write contracts with 2 December 2027 in mind now, because retrofitting is more expensive than choosing well from the start.
For help with the inventory and classification, our privacy, GDPR and EU AI Act consulting is billed at a day rate of €1,200 (incl. VAT). In AI consulting, a 90-minute initial session for €490 (incl. VAT) clarifies which AI use cases are worth pursuing and how to plan them in line with the AI Act from day one. Our consulting does not replace legal advice. Book a no-obligation call.
Sources
All accessed 6 November 2026.
- European Commission, AI Act policy page (timeline and Omnibus changes): https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
- AI Act Service Desk, implementation timeline after the Digital Omnibus: https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act
- AI Act, consolidated text (as at 27 July 2026), Articles 3, 4, 5, 6, 25, 26, 27, 50, 57, 96, 99, 111, 113 and Annex III: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-113 (other articles under the same path)
- Regulation (EU) 2026/1744 (AI Omnibus), Official Journal L of 24 July 2026: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ:L_202601744
- European Parliament, Legislative Observatory, procedure 2025/0359(COD) (AI Omnibus): https://oeil.secure.europarl.europa.eu/oeil/en/procedure-file?reference=2025%2F0359%28COD%29
- European Parliament, Legislative Observatory, procedure 2025/0360(COD) (Digital Omnibus, COM(2025) 837): https://oeil.secure.europarl.europa.eu/oeil/en/procedure-file?reference=2025%2F0360%28COD%29
- European Commission, press release on the political agreement of 7 May 2026: https://digital-strategy.ec.europa.eu/en/news/eu-agrees-simplify-ai-rules-boost-innovation-and-ban-nudification-apps-protect-citizens
- AI Act Service Desk, Digital Omnibus FAQ: https://ai-act-service-desk.ec.europa.eu/en/faq?faq_category_id=99
- AI Act Service Desk, draft guidelines on high-risk classification: https://ai-act-service-desk.ec.europa.eu/en/guidelines-classification-high-risk-ai-systems
- AI Act Service Desk, national resources: https://ai-act-service-desk.ec.europa.eu/en/national-resources
- European Commission, guidelines on Article 50 transparency obligations: https://digital-strategy.ec.europa.eu/en/policies/guidelines-transparency-ai-generated-content
- European Commission, Code of Practice on transparency of AI-generated content: https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content
- German Federal Law Gazette, BGBl. 2026 I No. 223 (AI Act implementing act): https://www.recht.bund.de/bgbl/1/2026/223/VO.html
- Future of Privacy Forum, "The AI Act Implementation Timeline: What Changes Under the AI Omnibus?" (28 July 2026): https://fpf.org/blog/the-ai-act-implementation-timeline-what-changes-under-the-ai-omnibus/
- Future of Life Institute, AI Act implementation timeline: https://artificialintelligenceact.eu/implementation-timeline/
Frequently asked questions
Has the EU AI Act applied in full since 2 August 2026?
No. Since 2 August 2026 the AI Act applies generally, including the Article 50 transparency rules, and national authorities and the AI Office enforce it. After the AI Omnibus, however, the high-risk obligations only apply from 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
What did the AI Omnibus change?
Regulation (EU) 2026/1744 has been in force since 27 July 2026. It postpones the high-risk obligations, gives providers of generative AI already on the market until 2 December 2026 to meet the Article 50(2) marking duty, and adds a new ban on AI that generates child sexual abuse material or non-consensual intimate content. It also softens the wording of the AI literacy duty and extends SME relief to small mid-caps.
Do SMEs still have to train staff on AI?
Yes. Article 4 has applied since 2 February 2025 and was not removed by the Omnibus. Companies must now take measures to support the development of AI literacy rather than guarantee a specific level for each person. Training, a short AI policy and records remain the practical way to show compliance.
Is a website chatbot a high-risk AI system?
Usually not. A website chatbot that answers questions or takes booking requests falls under the Article 50(1) transparency duty: people must be told they are interacting with AI. That duty has applied since 2 August 2026. A system becomes high-risk when, for example, it is used to select job applicants or assess people's creditworthiness.
Who enforces the AI Act in my country?
Each Member State designates its own market surveillance authorities, so the answer differs by country; the AI Office supervises general-purpose AI models at EU level. The Commission's AI Act Service Desk lists national resources and single points of contact. In Germany, for example, the Bundesnetzagentur is the central authority under a law in force since 29 July 2026.
Has the Digital Omnibus on the GDPR been adopted?
No. The separate proposal COM(2025) 837, which would amend the GDPR, the ePrivacy Directive and the Data Act among others, is still at committee stage in the European Parliament according to its Legislative Observatory. Until it is adopted, the GDPR applies unchanged.


