Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, entered into force on 27 July 2026. It amends the AI Act (Regulation (EU) 2024/1689), the aviation safety regulation and the machinery regulation. Everything in it is binding law. There is no guidance involved, which makes it simpler to read than most compliance news this summer.
What's affected
The regulation's title carries the date 8 July 2026. The new application dates, as listed on the AI Act Explorer:
| Date | What applies |
|---|---|
| 2 December 2026 | New Article 5 prohibitions on non-consensual intimate material and child sexual abuse material. Generative AI providers whose systems were placed on the market before 2 August 2026 must meet Article 50(2) marking by this date |
| 2 August 2027 | At least one national AI regulatory sandbox must be operational |
| 2 December 2027 | Annex III high-risk AI systems (Chapter III obligations) |
| 2 August 2028 | Annex I high-risk AI systems, meaning AI in regulated products |
| 2 August 2030 | High-risk systems used by public authorities |
Other changes in the text:
- AI literacy becomes a duty to support staff development, not a guarantee of a particular skill level.
- Providers may process special category personal data for bias detection, under strict safeguards.
- The definition of "safety component" is clarified, and non-safety uses are excluded.
- Documentation and quality management are simplified for SMEs and small mid-cap enterprises, with lower fine caps for them.
- The AI Office gets wider investigation, enforcement and fining powers over certain general-purpose AI based systems.
What did not move, per a Wilson Sonsini summary: Article 50 transparency duties have applied since 2 August 2026, and only the machine-readable marking duty has the grace period to 2 December 2026, and only for systems already on the market before 2 August 2026.
How to check if you're exposed
- List every AI feature you ship or embed, and note whether it falls under Annex III use cases such as employment, credit scoring or education, or sits inside an Annex I regulated product.
- For each, write the new date from the table above next to the old one, so planning documents stop quoting 2 August 2026.
- Mark features that generate synthetic audio, image, video or text. Article 50 is already live for them.
- Note whether any system was placed on the market before 2 August 2026. Only those get the marking grace period.
What to do now
- Update your internal roadmap and customer-facing statements to the new dates. Do not tell customers a high-risk feature is exempt today; it is deferred, not cancelled.
- Keep building the technical documentation. The obligations still arrive, 16 months or so later for Annex III.
- Decide how you will meet Article 50(2) marking for generative features before 2 December 2026 if they predate August.
- Check whether the SME and small mid-cap relief applies to you before you cut documentation work.
How suppliers can prepare
- Keep an inventory of models, datasets and AI components, so you can show which system is high-risk and which is not.
- Use the extra time for evidence, not delay. Risk management records, logging and human oversight design take longer than code.
- Treat the dates as a ceiling. Customers in regulated sectors may ask for conformity evidence before the legal deadline.
- Track the Commission's pending acts: delegated acts on limiting certain high-risk requirements and guidelines on complementarity with Annex I legislation are due by 2 August 2027.
Quick check
Is the Omnibus guidance or law?
Law. Regulation (EU) 2026/1744 is binding and amends the AI Act directly.
Which AI Act duty did the Omnibus not postpone?
Article 50 transparency duties apply from 2 August 2026. Only the Article 50(2) marking duty has a grace period, to 2 December 2026, for systems placed on the market before 2 August 2026.
Try it yourself
Click through it at your own pace, or open it full screen.
