Digital Omnibus Moves AI Act High-Risk Deadlines to December 2027 and August 2028
The Digital Omnibus on AI moves the EU AI Act's high-risk deadlines to December 2027 and August 2028, and adds a ban on sexual deepfakes and child abuse material.
The EU’s AI Act was written to take effect in stages. For high-risk AI systems, the first major date was 2 August 2026, and that date no longer applies to the stand-alone systems covered by the Act’s Annex III. An amending regulation has moved the deadlines and added a new prohibition.
The amendment is Regulation (EU) 2026/1744, known as the Digital Omnibus on AI. It amends the AI Act and two other EU regulations, and it is part of the bloc’s simplification agenda. The Council’s release places it in a package called Omnibus VII. Companies that place AI systems on the EU market, or use them there, now work from a different calendar.
A New Calendar for High-Risk Systems
Stand-alone high-risk AI systems now have an application date of 2 December 2027, instead of 2 August 2026. Cuatrecasas and Gibson Dunn both describe the original date as 2 August 2026.
High-risk systems embedded in regulated products, covered by Annex I, move to 2 August 2028. Cuatrecasas gives that date for systems under Annex I, section A.
The Council’s release, carried on the EEAS website, gives the same two dates. It describes the change as a replacement of the original 2 August 2026 date for stand-alone systems.
The two categories cover different kinds of systems. Gibson Dunn lists recruitment, credit scoring, education, law enforcement and border control as examples of stand-alone systems. It gives medical devices, machinery and vehicles as examples of AI embedded in regulated products.
Gibson Dunn says the fixed dates replace a conditional trigger mechanism that was part of the Commission’s proposal. The Parliament’s committees gave a reason for the delay: key standards may not be finalised by 2 August 2026. Their proposal set the same two dates that appear in the final text.
Cuatrecasas says units of a high-risk system lawfully placed on the market before the applicable date may continue to be placed on the market, provided the design is unchanged.
According to the Council’s release, products covered by the Machinery Regulation are exempt from direct application of the AI Act. The Commission can instead adopt secondary legislation under that regulation to add health and safety requirements for high-risk AI systems. Cuatrecasas says those delegated acts must be adopted by 2 August 2028 at the latest.
Official Journal Date
Regulation (EU) 2026/1744 amends the AI Act, Regulation (EU) 2024/1689, and was published in the Official Journal on 24 July 2026, according to Cuatrecasas.
Transparency Duties Stay on Schedule, With a Shorter Watermark Grace Period
The disclosure obligations under Article 50 keep their original start on 2 August 2026, according to Gibson Dunn.
The watermarking requirement for AI-generated content has a grace period for AI systems placed on the market before 2 August 2026. Gibson Dunn gives the end of that grace period as 2 December 2026. The Council’s release gives the same 2 December 2026 deadline and describes the change as a cut from six months to three, while Gibson Dunn and Cuatrecasas describe a four-month period.
A committee position adopted by Parliament’s IMCO and LIBE committees in March had proposed 2 November 2026 instead. The later date is the one given in the Council release and in the Gibson Dunn summary.
The earlier proposals were six months from the Commission, ending 2 February 2027, and three months from Parliament, ending 2 November 2026, according to Cuatrecasas.
The Council’s release says the rule covers transparency solutions for artificially generated content.
A New Prohibition on Sexual Deepfakes and Child Abuse Material
The amended Article 5 adds a prohibition covering non-consensual intimate deepfakes and AI-generated child sexual abuse material, according to Cuatrecasas.
The Council’s release says the ban covers AI practices that generate non-consensual sexual and intimate content, including child sexual abuse material. It says the ban is set to apply as of December 2026.
Gibson Dunn describes a transitional period for the new prohibition that runs until 2 December 2026.
Cuatrecasas says the ban covers intimate or sexually explicit depictions of an identifiable person. The consent it describes must be freely given, specific, informed and unambiguous.
For providers, the ban applies when generating such content is the intended purpose, or when a system’s design makes that output reasonably foreseeable and reproducible without adequate safety measures. For deployers, it applies only when they use a system specifically to generate or manipulate such material.
Cuatrecasas says altering existing material is not manipulation unless it increases exposure of intimate parts or changes sexually explicit activities.
For child sexual abuse material, Cuatrecasas says the ban has an exception where a defence is available under national law.
How the Deal Moved Through the EU Institutions
The Commission tabled the Digital Omnibus on AI on 19 November 2025, according to Gibson Dunn. The European Parliament’s press release gives the same date for the Commission’s proposal.
The European Parliament’s committees, IMCO and LIBE, adopted their joint position on 18 March 2026, by 101 votes to 9, with 8 abstentions, according to the Parliament’s press release. A plenary vote on Parliament’s mandate was expected on 26 March 2026, after which negotiations with the Council could begin.
Gibson Dunn says the first trilogue, on 28 April 2026, ended without agreement.
Gibson Dunn dates the provisional political agreement to 6 May 2026, and says member state representatives confirmed it on 13 May.
Gibson Dunn says formal adoption was expected before 2 August 2026, and that the changes take legal effect only on publication in the Official Journal.
The Council gave its final green light to the text on 30 June 2026, according to its release carried on the EEAS site.
Cuatrecasas gives 24 July 2026 as the date of publication in the Official Journal. It says the regulation enters into force three days after publication, which would be 27 July 2026.
Cuatrecasas gives the regulation’s date as 8 July 2026.
Stand-alone high-risk AI systems now have an application date of 2 December 2027, instead of 2 August 2026.
Key Dates for Compliance Teams
The main dates now run as follows: 2 August 2026 for the Article 50 disclosure duties, 2 December 2026 for the end of the new prohibition’s transitional period and the watermarking grace period, 2 August 2027 for member states’ AI regulatory sandboxes, 2 December 2027 for Annex III systems, and 2 August 2028 for Annex I systems.
The sandbox deadline is one year later than the original schedule, according to Gibson Dunn. The Council’s release gives the deadline for national competent authorities to establish AI regulatory sandboxes as 2 August 2027.
The AI Office will supervise AI systems built on general-purpose AI models when the model and the system come from the same provider, according to the Council’s release. Cuatrecasas describes that supervision as exclusive. Gibson Dunn says the AI Office also gets powers to investigate, carry out on-site inspections, accept binding commitments and impose fines.
Gibson Dunn says the AI Office’s competence also covers AI systems integrated into very large online platforms and very large online search engines under the Digital Services Act. National authorities keep competence in law enforcement, border management, judicial authorities and financial institutions.
What the Sources Say About Compliance Timing
Gibson Dunn says 2 August 2026 still matters, particularly for the Article 50 transparency duties. It advises starting or continuing compliance preparation rather than deferring it, even though the high-risk dates are later.
On AI literacy, Gibson Dunn says Article 4 was softened. Providers and deployers must support AI literacy among staff, rather than guarantee a specific level.
Gibson Dunn also says the legal basis for processing special category data to detect and correct bias now extends to all AI systems and general-purpose AI models, subject to a strict necessity standard.
For small mid-cap companies, Cuatrecasas says fines do not exceed the lower of the applicable percentages or amounts.




