Digital Omnibus AI Act: What Regulation 2026/1744 Changes

What is the Digital Omnibus on AI, and is it in force?

The "Digital Omnibus on AI" is Regulation (EU) 2026/1744, which entered into force on 27 July 2026 and amends the AI Act (Regulation (EU) 2024/1689) in a targeted way [39][26]. As of the date of this article, it therefore applies. The Commission presents it as a text that extends certain deadlines and introduces administrative simplification [2].

A simplification text, not an overhaul

According to the Commission, the amendments aim to simplify the implementation of AI legislation and keep rules clear, simple and conducive to innovation [3]. A specialist commentator summarises the objective as removing redundancies, clarifying obligations and easing the burden on certain businesses, particularly SMEs [18]. The text also amends Regulation (EU) 2018/1139 on aviation safety and Regulation (EU) 2023/1230 [10]. The risk-based approach, the foundation of the AI Act, is retained [21].

Not to be confused with the digital package

Regulation 2026/1744 covers only the AI component of a broader set of measures. On 19 November 2025, the Commission presented a simplification package proposing to amend the AI Act, the GDPR, the Data Act and other digital texts at the same time [19]. The AI-specific proposal, called the "AI omnibus", was adopted on that date. A political agreement was reached on 7 May 2026, and the text then entered into force on 27 July [3]. The European Parliament had voted on it on 16 June 2026 [26]. Changes to the GDPR and the Data Act belong to other parts of this package, which are not covered here.

What the sources allow us to state

The primary sources in the file, issued by the Commission, confirm the entry into force on 27 July 2026 and the principle of a phased timetable [1][2][3]. The details of the measures (dates by category of system, relief for SMEs, added prohibitions) come mostly from secondary sources: law firms, publishers and specialist press. They are addressed in the following sections, with the nature of each source noted each time. Only the text published in the Official Journal of the European Union is authoritative, and the assessment of compliance remains the responsibility of each company and its advisers.

What is the adoption timeline of the text?

The Digital Omnibus on AI was adopted in five stages, from November 2025 to July 2026, and Regulation (EU) 2026/1744 entered into force on 27 July 2026 [3][26].

• 19 November 2025: Commission proposal. The Commission publishes a simplification package proposing to amend the AI Act, the GDPR, the Data Act and other digital texts simultaneously [19]. It refers to the AI component as the "AI omnibus" [3].
• 7 May 2026: political agreement. The Council and the European Parliament reach a provisional political agreement on the Digital Omnibus on AI [11]. The Commission confirms this date [3].
• 16 June 2026: European Parliament vote. Parliament adopts the text, which amends the application timetable of the AI Act [7][26]. Another source confirms a vote in June 2026 [27].
• 8 July 2026: date of the regulation. A secondary source cites Regulation (EU) 2026/1744 as "of 8 July" [5]. The file contains no primary source confirming this date, which must therefore be checked against the text published in the Official Journal of the European Union.
• 27 July 2026: entry into force. The Commission states that the AI omnibus enters into force on that day throughout the EU [3]. A secondary source gives the same date [26].

One source contradicts this timeline

An article claims that the Digital Omnibus was "adopted in March 2026" [29]. This date is consistent neither with the Commission, which places the political agreement in May 2026 [3], nor with the sources that place the Parliament's vote in June 2026 [7][26]. This article is a secondary commercial source. We retain the timeline of the Commission and Parliament.

For legal purposes, the dates to cite are those of the official text. Their interpretation, like their effects on the company, is for the company and its advisers to assess.

Which "high-risk" obligations are postponed, and to when?

The obligations applicable to high-risk AI systems in Annex III of the AI Act, originally due on 2 August 2026, will ultimately apply from 2 December 2027 [20][30]. This is the central postponement introduced by Regulation (EU) 2026/1744, which entered into force on 27 July 2026 [2][26]. The gap between the two dates is about 16 months.

A second deadline to verify

Only one source in the file mentions 2 August 2028: its extract breaks off after "pass to 2 December 2027 and 2 August 2028" [26]. The attachment of this date to Annex I, that is, to products already subject to sector-specific regulation, therefore does not appear in the available extracts. The Commission's pages on the omnibus [2][3] do not give the details either, in the extracts consulted. This date should not be carried over as it stands into a compliance plan before reading the official text of Regulation (EU) 2026/1744.

Why these requirements

According to one commentary, the regulation eases documentation burdens and leaves "between sixteen and twenty-four additional months on the most costly requirements" [38]. Another cites, among the systems concerned, those deployed in sensitive sectors such as critical infrastructure and biometrics [4]. Both sources are third-party analyses, not official texts.

This timetable is an interpretation of the text. Whether a system falls under Annex III, and by what deadline, is an assessment for the company and its advisers.

What already applies despite the postponement?

Four blocks of AI Act obligations remain due despite the postponement of the "high-risk" rules: prohibited practices, AI literacy (Article 4), general-purpose AI models, and the transparency requirements of Article 50 [34].

• Prohibited practices: applicable since 2 February 2025 [34].
• AI literacy (Article 4): maintained, in a streamlined version [14].
• General-purpose AI (GPAI) models: obligations due since 2 August 2025 [14].
• Transparency (Article 50): applicable since 2 August 2026 [17][13].

The transitional period for labelling

Regulation 2026/1744 provides an additional period until 2 December 2026 for labelling AI-generated content [16]. It targets systems already on the market before 2 August 2026, i.e. a four-month transition [14]. The sources phrase the reference date slightly differently ("before this date" in [16], "before 2 August 2026" in [14]). For a system placed on the market close to that date, the text itself must therefore be checked.

What the Article 4 relief contains

The sources in the file describe Article 4 as "eased" [37] or "streamlined" [14], but they do not detail its content. None specifies which wording was removed, modified or added. This article therefore does not paraphrase them: the exact scope of the relief can be read in the text of Regulation (EU) 2026/1744, by comparison with the original Article 4. Until that reading is done, it would be imprudent to consider the AI literacy obligation as removed or emptied of substance.

A postponement that changes nothing for most

One source notes that, for most French businesses, whose AI uses are ordinary rather than high-risk, the AI Act applies now and the postponement changes nothing for their situation [40]. This view comes from a commercial player. It is consistent with the scope of the postponement, which concerns the Annex III obligations [13]. The classification of each use and the assessment of compliance remain those of the company and its advisers.

What simplifications are there for SMEs and "small mid-caps"?

Regulation (EU) 2026/1744 extends to small mid-caps the simplification measures previously reserved for SMEs [37]. According to a law firm, it also introduces a formal definition of SMEs and small mid-caps in the regulation [15]. The sources consulted do not reproduce the quantitative criteria of this definition. They must therefore be checked in the official text.

The measures cited by the sources

• Simplified technical documentation for high-risk systems [15][25]. Another source refers to simplified forms [14].
• Proportionate quality management requirements [15].
• Priority access to national regulatory sandboxes and to the EU's central sandbox [25].
• Proportionate fine caps [25], described elsewhere as "reduced" [15]. The exact method of calculation does not appear in the available extracts.
• Removal of two formalities [37]. The sources do not identify them.

These relief measures come from law firm analyses and specialist websites, not from the text of the regulation itself. They must be checked against its published version.

The case of Article 4a

One source mentions a new Article 4a. It specifies that it "is not an obligation" but "an optional legal basis" [14]. The file does not detail its purpose or conditions of use, so this article should not be treated as an additional requirement.

A critical viewpoint

These simplifications are not uncontroversial. The French Senate published a document entitled "European digital omnibus: a risk for the…" [36]. It notes that several measures aim to simplify administrative obligations for SMEs and small mid-caps [36]. The available extract is truncated, so its critical arguments cannot be detailed here. Reading it in full is useful for lawyers and executives who want to weigh the risks associated with simplification.

Key takeaway for an affected company

Whether a company falls within these measures depends on the definition laid down by the regulation. Whether it does, and what that changes for its own systems, is for the company and its advisers to assess.

What new prohibitions does the Digital Omnibus add, and when do they apply?

The Digital Omnibus adds two new prohibitions to Article 5 of the AI Act, which lists prohibited practices, but the sources in the file differ on their application date. Article 5 has applied since 2 February 2025 [34]; Regulation (EU) 2026/1744 therefore extends its list.

The two added prohibitions

According to the sources consulted, the new prohibitions target AI systems that generate:

• non-consensual sexually explicit intimate content [10];
• child sexual abuse material [10].

Unlike the postponement of "high-risk" obligations, these additions widen the scope of what is prohibited rather than easing it [7].

A divergence on the application date

Two sources give different indications:

The file does not allow us to decide between these two versions. The available extracts are partial, and none reproduces the article of the regulation that sets this date. Nor should the two dates be confused: 2 December 2026 is also the date of the additional period for labelling AI-generated content already on the market [16]. That concerns a different obligation, relating to transparency.

Check the official text

The exact date must be read in Regulation (EU) 2026/1744 itself, published in the Official Journal of the European Union, which alone is authoritative. This regulation entered into force on 27 July 2026 [2][26]. The provision setting the date of application of the new prohibitions is to be found there.

In the meantime, a company that develops, integrates or deploys tools for generating images, videos or content would be well advised to treat these uses as prohibited without waiting for the latest possible date. Checking its tools, their settings and the uses actually open to its teams is a matter of common sense. The assessment of compliance remains that of the company and its advisers.

Is the postponement a general pause in compliance?

No, the postponement is not a general pause in compliance. It shifts a specific deadline, that of the Annex III "high-risk" obligations, from 2 August 2026 to 2 December 2027 [30]. According to practitioners' analyses, it "does not suspend obligations already in force nor alter the substance of the requirements: classification, technical documentation, FRIA, human oversight" [22].

What Regulation 2026/1744 does not touch

• The tiered risk architecture. The text does not modify it, but it shifts the obligations for high-risk systems and adds new deadlines [28].
• The risk-based approach. It remains the foundation of European AI regulation [21].
• The content of the requirements. Systems must still be classified, documented, subject to human oversight and, where applicable, subjected to a fundamental rights impact assessment (FRIA) [22].

The postponement changes when these requirements become enforceable, not their nature. Companies that have not started their inventory of AI systems therefore remain exposed to a preparation lag [22]. According to another analysis, the regulation "eases documentation burdens" and leaves sixteen to twenty-four additional months on the most costly requirements. That extra time mainly benefits those that needed time to document their systems [38].

Obligations already enforceable remain so

For most French businesses, whose AI uses are ordinary rather than high-risk, the AI Act applies now and the postponement changes nothing for their situation [40]. This covers the prohibitions, AI literacy (Article 4, in its streamlined version) and the transparency requirements of Article 50 since 2 August 2026 [14][17][34].

An ongoing litigation risk

The postponement does not "wipe out" litigation risk: the penalties provided for by the AI Act remain applicable to obligations already enforceable [30][40]. The exact amounts are to be read in the text of Regulation (EU) 2024/1689 as amended, not in secondary summaries.

Most of these analyses come from law firms and consultants. They help in understanding the text but do not replace reading it or obtaining legal advice. The assessment of an AI system's compliance is the responsibility of the company and its advisers.

What to do concretely between now and December 2027, and how can Oupi help?

The most useful approach between now and December 2027 is to proceed step by step: inventory, classify, verify what is already enforceable, then plan for what is postponed. Regulation (EU) 2026/1744 has been in force since 27 July 2026 [2][3]. It pushes the obligations for Annex III high-risk systems back to 2 December 2027 [30]. This postponement "does not suspend obligations already in force nor alter the substance of the requirements" [22].

1. Inventory your AI uses. List the systems used or provided, including those introduced without going through the IT department. The inventory is the prerequisite for any classification [22].
2. Classify by risk level. The Digital Omnibus retains the risk-based approach [21]. Classification is part of the requirements whose substance remains unchanged [22].
3. Verify what is already enforceable.
   • Article 4 on AI literacy, in its streamlined version [14][37].
   • The transparency requirements of Article 50, applicable since 2 August 2026, with a four-month transition for systems already on the market before that date [14][17].
   • The obligations relating to general-purpose AI models, due since 2 August 2025 [14].
4. Plan for the requirements due in December 2027. These include technical documentation, the fundamental rights impact assessment (FRIA) and human oversight [22]. Some obligations are tied to a 2 August 2028 deadline [26]. Also check whether your company falls within the simplifications extended to small mid-caps [37].

Where Oupi can help

Oupi monitors how the text evolves and helps document the measures taken, in particular training and traceability of uses. This support produces evidence and monitoring records. It is not a promise of compliance.

A framework to keep in mind

Only the official text is authoritative. The European Commission's pages [2][3] and the regulation published in the Official Journal of the European Union are the reference. The legal qualification of each system, and therefore the assessment of compliance, rests with the company and its advisers [23].