Agent Liability Europe

The Digital Omnibus, provision by provision.

The Digital Omnibus on AI entered into force on 27 July 2026, six days before the original high-risk deadline. Annex III high-risk obligations now apply from 2 December 2027 and Annex I from 2 August 2028. This page records what moved, what did not, and what binds a deployer today. The legislative passage is closed and kept below as the record.

Every date here was re-verified on 17 August 2026 against the European Commission and the Council of the EU. Where a detail could not be read at an institutional source, the row says so rather than guess. No carrier or vendor pays for placement on this tracker.

In forceSince 27 July 2026, as Regulation (EU) 2026/1744.
0Days until Annex III applies, on 2 December 2027. Stand-alone high-risk AI systems.
0Days until Annex I applies, on 2 August 2028. High-risk AI embedded in products.
Art. 4, 5, 50, 99Already in application: literacy and prohibitions since 2 February 2025, GPAI since 2 August 2025, transparency and penalties since 2 August 2026.

The dates still ahead.

The legislative watch is over. What replaces it is a compliance calendar, set by the AI Act as amended and verified against the Commission's own implementation timeline on 17 August 2026.

  1. The Omnibus in forceRegulation (EU) 2026/1744 amends the AI Act.
  2. Transparency and enforcementArticle 50 applies. Penalties apply to everything in application.
  3. Article 50(2) marking, and the new prohibitionGenerative systems already on the market must mark synthetic content, and the ban on CSAM and non-consensual intimate imagery reaches them.
  4. National AI regulatory sandboxesAt least one sandbox running in each Member State. For a deployer, the signal of when national authorities intend to be ready.
  5. Annex III high-risk obligationsThe full high-risk regime, the Article 26 deployer file and the Article 27 FRIA.
  6. Annex I high-risk obligationsAI embedded in products under EU harmonisation legislation, such as medical devices and machinery.

A note on the gap. The time before 2 December 2027 does not mean the work went away. Harmonised standards were still unpublished at the last verification, and the deferral exists because of that. An operator who uses the time to build the operator file against the current text will be ready. An operator who waits for the standards will be doing the work in the last quarter of 2027 alongside everyone else.

Who each date reaches. 2 December 2026: anyone shipping a generative image, audio or video product in the EU, and platforms hosting them. 2 December 2027: deployers and providers of employment, credit, insurance pricing, education, essential services, law enforcement and migration systems, for risk management, data governance, technical documentation, logging, human oversight, accuracy and robustness. 2 August 2028: manufacturers of medical devices, machinery, lifts, toys and the rest of the Annex I list, and their notified bodies.

The provision matrix.

Each row gives the date the AI Act originally set, the date that applies after the Omnibus, and how the provision came out. Click a column heading to sort, or use the filters to narrow the view.

Show
Provision Original date Date now Status Council Parliament Outcome Notes Source

What applies right now.

The deferral moved two things, Annex III and Annex I. Everything below was outside its scope and is in application and enforceable today. It is the list an operator most often gets wrong, because the headline said the AI Act was delayed and only one part of it was.

  1. 01

    Article 5 prohibitions

    Since 2 February 2025

    The prohibited AI practices were untouched by the deferral. Enforcement of them began on 2 August 2026. A system using a prohibited practice must stop. Source: Regulation (EU) 2024/1689, Article 5 and Article 113(1)

  2. 02

    Article 50(1) chatbot disclosure

    Since 2 August 2026

    A deployer of an AI system that interacts with a natural person must tell that person they are dealing with an AI, clearly and in time for it to matter. Not deferred. Source: Regulation (EU) 2024/1689, Article 50(1)

  3. 03

    Article 50(3) emotion recognition and biometric categorisation

    Since 2 August 2026

    A deployer of a system that infers emotion or categorises a person biometrically must inform the people exposed to it. Not deferred. Source: Regulation (EU) 2024/1689, Article 50(3)

  4. 04

    Article 50(4) deepfake labelling

    Since 2 August 2026

    Anyone using AI to generate or manipulate image, audio or video that constitutes a deepfake must disclose it clearly and visibly. Not deferred. Source: Regulation (EU) 2024/1689, Article 50(4)

  5. 05

    Article 50(2) machine readable marking

    Transition ends 2 December 2026

    A provider of a synthetic content system already on the market before 2 August 2026 has until 2 December 2026 to meet the machine readable marking requirement. It is the next binding date on this page. Source: EC AI Act Service Desk implementation timeline. Verified 17 August 2026.

  6. 06

    The new prohibition on non-consensual intimate imagery and CSAM generation

    Existing systems by 2 December 2026

    Added by the Omnibus. Systems already on the market have until 2 December 2026 to comply. Source: EC AI Act Service Desk implementation timeline. Verified 17 August 2026.

  7. 07

    GPAI obligations, Articles 53 and 55

    Since 2 August 2025

    Transparency, copyright policy and systemic risk requirements for general purpose AI models. Not deferred. Enforcement began on 2 August 2026. Source: Regulation (EU) 2024/1689, Articles 53, 55 and 113(2)

  8. 08

    Article 4 AI literacy, as amended

    Since 2 February 2025

    Still an obligation on providers and deployers, with no mandated level. For a deployer of a high-risk system, the duty to train staff so that human oversight is real remains. Enforcement began on 2 August 2026 and sits with national market surveillance authorities. Source: EC AI literacy questions and answers. Verified 17 August 2026.

  9. 09

    Article 99 penalties

    Since 2 August 2026

    The penalty regime applies to obligations that are themselves in application. Everything in this list is enforceable now. Annex III duties are not, until 2 December 2027. Source: Regulation (EU) 2024/1689, Article 99 and Article 113(3)

The passage, a closed file.

The Omnibus followed the ordinary legislative procedure under Article 294 TFEU, in six stages from the Commission proposal on 19 November 2025 to entry into force on 27 July 2026.

  1. Commission proposalCOM(2025) 836.
  2. Council and Parliament positionsExact dates not confirmed at an institutional source.
  3. Political agreementThe new dates fixed.
  4. Parliament plenary voteWith the new prohibition.
  5. Council final adoptionThe last legislative act.
  6. Entry into forceRegulation (EU) 2026/1744.
  1. 1

    Commission proposal

    19 November 2025

    The European Commission presented COM(2025) 836, the Digital Omnibus on AI, a targeted amendment to Regulation (EU) 2024/1689. It proposed linking the application of the high-risk rules to the availability of harmonised standards and supporting tools, with a backstop of 2 December 2027 for stand-alone high-risk AI systems and 2 August 2028 for high-risk AI embedded in products. It travelled alongside COM(2025) 834, the broader Digital Omnibus touching the GDPR, the Data Act and NIS 2. Source: European Commission, Digital Omnibus on AI Regulation Proposal. Verified 17 August 2026.

  2. 2

    Council and Parliament negotiating positions

    Spring 2026

    The Council adopted a general approach and the European Parliament a negotiating mandate on the basis of a joint IMCO and LIBE committee position. Both broadly accepted a deferral of the stand-alone high-risk obligations. We previously published specific committee and plenary dates for this stage. They could not be confirmed against a Council or Parliament document on 17 August 2026 and have been withdrawn rather than restated. Source: European Parliament Legislative Train, Digital Omnibus on AI

  3. 3

    Political agreement

    7 May 2026

    Council and Parliament negotiators reached a provisional political agreement. The open question through the negotiation had been whether the new dates would be fixed or conditional on standards being ready. The co-legislators fixed them: 2 December 2027 and 2 August 2028, with no condition tied to the arrival of standards. Source: Council of the EU press release, 7 May 2026

  4. 4

    European Parliament plenary vote

    June 2026

    The Parliament approved the agreed text in plenary, together with the new prohibition on AI systems that generate child sexual abuse material or that depict an identifiable person's intimate parts without consent. Systems already on the market have until 2 December 2026 to comply with that prohibition. Source: European Parliament press release. Verified 17 August 2026.

  5. 5

    Council final adoption

    29 June 2026

    The Council gave final approval to the regulation simplifying the implementation of the harmonised rules on artificial intelligence, the last legislative act required before publication in the Official Journal. Source: Council of the EU press release, 29 June 2026

  6. 6

    Official Journal publication and entry into force

    27 July 2026

    The amending act is Regulation (EU) 2026/1744, published at OJ L, 2026/1744 on 24 July 2026 and in force on the third day after publication, 27 July 2026, six days before the original 2 August 2026 high-risk deadline. Anyone who planned on the original date is not in breach; the obligations simply have not started. Anyone who stood down a compliance programme on the assumption of a general reprieve should read the matrix above, because four sets of duties were never in scope of the deferral. Source: European Commission, AI Omnibus enters into force. Verified 17 August 2026.

How the tracker is kept.

How it is sourced, verified and updated.

Re-verified at source, on each material development

Every date on this page was read against the European Commission and the Council of the EU on 17 August 2026.

A tracker that is not re-read is a liability, and this one had been running on April 2026 facts before that pass. An Official Journal entry, a Commission implementation update or a Member State designation triggers a new pass. The date given is the last verification, not the last text edit.

Institutional sources only

The EC AI Act Service Desk implementation timeline; the Commission's Shaping Europe's digital future; Council of the EU and European Parliament press releases and the Legislative Train; EUR-Lex and the Official Journal; Member State ministries, regulators and official gazettes.

Law firm trackers and news reporting help find a document. They never confirm one. Neither does a search engine summary, which in this subject now reflects secondary commentary back as if it were the source.

Read at the issuing body, or marked as unread

Every date was read on the website of the institution that owns it. Where that was not possible in the 17 August 2026 pass, the row says so in plain words and offers no number.

Two rows say it: the Council and Parliament position dates in the passage record, and the amended Article 71 registration mechanics. We would rather show a gap than fill one.

No carrier or vendor pays for visibility

No insurer, AI vendor, law firm or other commercial party pays for inclusion, priority or framing here. Coverage decisions are made by the editors on legal materiality alone.

The tracker is published under CC BY 4.0. You may reproduce it with attribution.

Last verified17 August 2026 Next binding date2 December 2026 LicenceCC BY 4.0
The master brief

What the amendment changed, and what to do with the time.

What the Omnibus changed, what it left alone, and what a deployer should be doing before Annex III applies on 2 December 2027.