What the Omnibus changed for engineering teams (a diff, not a deadline list)
AI Act high-risk delay to December 2027: what to do now. A diff of what Regulation (EU) 2026/1744 changed, for engineering teams.
Regulation (EU) 2026/1744, the Digital Omnibus on AI, amends the EU AI Act for AI providers and deployers in the EU. Deployers are companies that use an AI system under their own authority. It entered into force on 27 July 2026 (Art 4 of the Omnibus). Annex III high-risk duties apply from 2 December 2027, and Annex I product duties from 2 August 2028 (Omnibus Art 1(40), amending Art 113). Annex III is the Act's list of high-risk uses, and Annex I lists EU product-safety laws. Annex IV documentation is the technical documentation that Annex IV requires for high-risk systems. Two new bans cover AI systems that generate or manipulate non-consensual intimate images or child sexual abuse material (Omnibus Art 1(7), amending Art 5). They apply from 2 December 2026. Non-compliance with an Art 5 ban is subject to administrative fines (Art 99(3)). Fines reach up to EUR 35 000 000 or, for an undertaking, up to 7 % of worldwide annual turnover, whichever is higher (Art 99(3)). For an undertaking, the 7 % applies to its total worldwide annual turnover of the preceding financial year (Art 99(3)). For a small or medium-sized enterprise (SME), including a start-up, each such fine is capped at the lower of the two limits (Art 99(6)). Providers who judge an Annex III system not high-risk under Art 6(3) must still register it before placing it on the market (Art 49(2)). The Omnibus keeps that duty and removes two data fields (Omnibus Art 1(42)). The Omnibus names only Chapter III, Sections 1 to 3 in its new application dates for high-risk duties (Omnibus Art 1(40)). Art 49 sits in Section 5 of Chapter III, so check with counsel from when Art 49(2) applies to your system. Re-baseline your roadmap to the Annex III and Annex I dates, keep building Annex IV documentation, and check image and video features against the new bans. Omnibus Art 1(n) means point (n) of Art 1 of the Omnibus. Plain article numbers belong to the AI Act, Regulation (EU) 2024/1689, as amended; Art 4a, 5(1a), 5(1b) and 111(4) exist only after the amendment.
Three headline dates moved later, and two duties apply from 2 December 2026
Five headline changes carry a date, and the general application date of 2 August 2026 did not change (Omnibus Art 1(40)). The Omnibus moves other dates too, such as the Art 72(3) template date, covered below.
| Change | Before | After | Source |
|---|---|---|---|
| Annex III high-risk duties | Date moved from 2 August 2026 | Apply from 2 December 2027 | Omnibus Art 1(40) (amending Art 113) |
| Annex I product duties | Date moved from 2 August 2027 | Apply from 2 August 2028 | Omnibus Art 1(40) (amending Art 113) |
| National AI regulatory sandbox | Date moved from 2 August 2026 | Operational by 2 August 2027 | Omnibus Art 1(22) (amending Art 57(1)) |
| Two new bans, Art 5(1)(ba) and (bb) | Did not exist | Apply from 2 December 2026 | Omnibus Art 1(40) (amending Art 113) |
| Art 50(2) marking, systems placed on the market before 2 August 2026 | No grace rule | Comply by 2 December 2026 | Omnibus Art 1(39) (inserting Art 111(4)) |
The full date table, with every other milestone, sits on EU AI Act Deadlines: Timeline From 2024 to 2030. The Omnibus does not amend Annex IV. The post-market monitoring plan is part of the Annex IV documentation (Omnibus Art 1(30), amending Art 72(3)). The Commission must adopt guidance, including a template, on that plan by 2 September 2027 (Art 72(3)).
Two new bans mean a check of every image and video feature
Check every image and video feature against two new bans that apply from 2 December 2026. Omnibus Art 1(7) adds them to Art 5(1) as points (ba) and (bb), and Omnibus Art 1(40) sets the date. Point (ba) bans AI systems that generate or manipulate realistic intimate images, video or audio of an identifiable person without that person's explicit consent. Point (bb) bans AI systems that generate or manipulate material or performance covered by Article 2, points (c) and (e), of Directive 2011/93/EU. For point (bb), the ban gives way where a "without right" defence applies under national law.
Art 5(1a) limits the ban on placing on the market and putting into service to two cases. In the first case, generating that material is the intended purpose of the system. In the second case, the system's design, training, architecture, capabilities or user-facing functions make that output reasonably foreseeable and reproducible. That must hold without significant technical modification. The second case also needs a system without reasonable and adequate technical safety measures to prevent it reliably and to correct reported misuse. Use is banned only where the deployer uses the system to generate or manipulate such material.
Art 5(1b) applies to point (ba) only. It says a modification that neither increases the exposure of depicted intimate parts nor alters the nature of depicted sexual activity is not manipulation. The AI Act, as amended, does not define "reasonable and adequate technical safety measures". Document the measures you use. If you ship an image or video generator, test it against both cases.
Registration for self-assessed systems still exists, with less data
The Omnibus does not amend Art 49 (registration) or Art 6(3): the Omnibus has no point that changes either. Art 6(3) is the route by which a provider concludes that an Annex III system is not high-risk. Omnibus Art 1(42) deletes points 7 and 9 of Annex VIII, section B, the data a provider submits for that registration. The deleted points covered the short summary of why the system is not high-risk and the list of Member States where it is available. The Omnibus names only Chapter III, Sections 1 to 3 in its new application dates for high-risk duties (Omnibus Art 1(40)). Art 49 sits in Section 5 of Chapter III, so check with counsel from when Art 49(2) applies to your system. Keep the evidence behind every self-assessment, and expect to submit fewer Annex VIII data fields.
AI literacy stays your duty, with softer wording
Art 4 still binds providers and deployers of AI systems (Omnibus Art 1(5), amending Art 4). Art 3(56) defines AI literacy as the skills, knowledge and understanding that allow an informed deployment of AI systems and awareness of their risks and possible harm. The original Art 4 wording asked them to ensure a sufficient level of AI literacy, to their best extent. The new wording asks them to take measures to support the development of AI literacy. That covers their staff and others who operate or use AI systems on their behalf. Measures take into account technical knowledge, experience, education, training and the context of use. Art 4 does not require providers or deployers to guarantee any specific level of AI literacy of any individual. The Commission and Member States must support providers and deployers, in particular SMEs. The Commission must also publish practical examples. Keep your literacy measures and record who received which one.
Bias-detection data rules moved from original Art 10(5) to a new Art 4a
Omnibus Art 1(9) deletes original Art 10(5), and Omnibus Art 1(6) inserts Art 4a, which holds the rule on special categories of personal data for bias detection. Special categories are sensitive data, such as health or ethnic origin. Providers of high-risk AI systems may exceptionally process them to the extent strictly necessary to detect and correct bias (Art 4a(1)). Art 4a(1) sets six conditions:
- Show that no effective alternative exists, including synthetic or anonymised data.
- Apply technical limits to re-use of the data.
- Apply access controls.
- Do not transmit the data to other parties.
- Delete the data once the bias is corrected or the retention period ends.
- Keep records of processing that state why the processing was strictly necessary.
Art 4a(2) lets providers and deployers of other AI systems and models do the same. Deployers of high-risk systems may do so too. Both need strict necessity and all the safeguards of paragraph 1. Art 4a(2) creates no duty to run bias detection. If your pipeline processes such data for bias work, check each condition and record the reasons.
Small mid-caps join SMEs for documentation, quality-system and penalty relief
The Omnibus adds the small mid-cap enterprise (SMC) to the Act (Omnibus Art 1(4), inserting Art 3(14a) and (14b)). An SMC is defined in point (2) of the Annex to Recommendation (EU) 2025/1099.
- Art 11(1): SMEs, including start-ups, and SMCs may provide Annex IV documentation in a simplified manner.
- Art 17(2): quality-management measures scale with organisation size, especially for SMEs and SMCs.
- Art 17(2): the required rigour stays.
- Art 99(1): Member States consider the interests and economic viability of SMEs, start-ups and SMCs when imposing penalties.
- Art 99(6a): SMC fines under paragraphs 4 and 5 are capped at the lower of the percentage or the amount.
- Art 63(1): SMEs, including start-ups, without partner or linked enterprises may simplify parts of the quality-management system.
- Art 63(1) does not name SMCs.
The Omnibus makes these changes in Omnibus Art 1(10), 1(11), 1(26) and 1(38). Check whether your company meets the Recommendation's SMC definition before you plan on the relief.
Marking synthetic output has a later deadline for systems already on the market
Providers of systems placed on the market before 2 August 2026 must comply with Art 50(2) by 2 December 2026 (Omnibus Art 1(39), inserting Art 111(4)). The rule covers providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text. Art 50(2) is the Act's duty to mark synthetic output. The Omnibus changes only paragraph 7 of Art 50 (Omnibus Art 1(20)); the marking duty in Art 50(2) keeps its text.
The AI Office supervises two groups of systems, and Art 25 changes value-chain duties
The Commission's AI Office alone supervises and enforces AI systems built on a general-purpose AI model when model and system share one provider (Omnibus Art 1(31), amending Art 75(1)). The article lists exceptions, among them Annex I products. The AI Office also supervises systems that are, or are integrated into, very large online platforms or search engines (Art 75(1)). Art 75a gives it market-surveillance powers, including information requests and remote and on-site inspections (Omnibus Art 1(32)).
Omnibus Art 1(12) amends Art 25. Under Art 25(2), the initial provider is no longer considered a provider of that system and must cooperate with the new provider. That rule is not new. The new part lists what the cooperation covers: technical documentation, information on limitations and failure modes, and targeted technical access. Art 25(4) calls for a written agreement with third parties that supply AI systems, models or tools, based on the generally acknowledged state of the art. AI models are newly named in that list. Art 25(4) does not apply to free and open-source tools other than general-purpose AI models. Art 99(4)(da) adds duties under Art 25(2) and (4) to the infringements listed in Art 99(4) (Omnibus Art 1(38)). Check your supplier contracts against Art 25(4).
What to do after the Omnibus: eight actions for engineering teams
- Re-baseline the roadmap: Annex III work to 2 December 2027, Annex I work to 2 August 2028 (Omnibus Art 1(40)).
- Keep building Annex IV documentation; the Omnibus does not amend Annex IV.
- Check image and video features against Art 5(1)(ba), (bb) and (1a).
- Keep the evidence behind each Art 6(3) self-assessment, and prepare the reduced Annex VIII data for registration (Art 49(2)).
- Keep your Art 4 literacy measures running and recorded.
- Mark synthetic output as Art 50(2) requires.
- Test any special-category data use for bias work against the six Art 4a conditions.
- Check your SMC status under Recommendation (EU) 2025/1099.
How Scanara helps with Annex IV documentation
Scanara produces Annex IV documentation aligned to the Annex IV template, from your inputs and your assessment answers. It does not make that documentation accurate or complete. You review it, correct it and confirm it. Scanara reduces the effort and error rate of building that documentation. Scanara also structures your risk classification and the Art 6(3) assessment, and you decide the outcome. Scanara structures registration evidence, and filing stays your act. Register your AI system in Scanara and answer the assessment to start your Annex IV draft.
Frequently Asked Questions
Did the Omnibus drop registration for self-assessed not-high-risk systems?
No. The Omnibus does not amend Art 49 or Art 6(3), and it deletes only points 7 and 9 of Annex VIII, section B (Omnibus Art 1(42)).
When do high-risk duties apply after the Omnibus?
Annex III high-risk duties apply from 2 December 2027 (Omnibus Art 1(40), amending Art 113). Annex I duties apply from 2 August 2028 (Omnibus Art 1(40)).
Do the new bans cover my image or video generator?
It can. The ban applies when generating the banned material is the intended purpose of the system (Art 5(1a)). It also applies when the design makes that output foreseeable and reproducible and the system lacks reasonable and adequate safety measures (Art 5(1a)).
Does Art 4 still apply to my team?
Yes. Art 4 still binds providers and deployers, who must take measures to support the development of AI literacy (Omnibus Art 1(5)). It does not require them to guarantee a specific level of AI literacy of any individual (Omnibus Art 1(5)).
Sources & References
- Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the implementation of harmonised rules on artificial intelligence (Digital Omnibus on AI, "the Omnibus"), OJ L, 24.7.2026, ELI: https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng
- Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 (Artificial Intelligence Act, "the AI Act"), OJ L, 12.7.2024, ELI: https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- Commission Recommendation (EU) 2025/1099 (definition of small mid-cap enterprise), OJ L, 2025/1099, 28.5.2025, ELI: http://data.europa.eu/eli/reco/2025/1099/oj
- Directive 2011/93/EU (Article 2, points (c) and (e)), OJ L 335, 17.12.2011, p. 1, ELI: http://data.europa.eu/eli/dir/2011/93/oj
Scanara prepares evidence from your repositories
Scanara reads your connected GitHub repositories and runs rule-based scans for standard AI and machine-learning patterns. It can miss non-standard code. Separately, it generates draft documentation from templates and your assessment answers as evidence for your compliance work. You make the classification decisions.