Analytical Assessment of Regulatory Realignment and Enforcement Pragmatism: The August 2026 Milestone Under the EU AI Act

This piece is authored by Sankalp Mirani, Legal Advisor at Examplad Media and Samiksa Shah, a Fifth year law student at Balasaheb Apte College of Law.

Today, 2 August 2026, is a turning-point in the European Union ambitious project to regulate artificial intelligence. Originally, in the proposed legislative timeline of the Regulation (EU) 2024/1689 ("AI Act"),[1] this date was intended as a complete implementation of the transparency requirements and obligations regarding high risk AI systems.

But things are very different in August 2026 when the regulation takes effect. EU institutions have made a pragmatic and structural adjustment of the compliance programme with the new Regulation (EU) 2026/1744 ("Digital AI Omnibus

") which came into force on 27 July 2026.[2]

This analysis focuses on the legal, institutional and strategic implications of this realignment. It examines the conflict between fast-track lawmaking and slow-track government, sheds light on the duties that will be enforced now, and considers the consequences of the indefinitely postponed compliance dates.

The Institutional Deficit and Genesis of the Digital AI Omnibus

The adoption of the Digital AI Omnibus[3] took place shortly before the 2 August deadline, and signals a tacit admission by the EU co-legislators of a lack of institutional and technical capacity. The original 2024 text of the AI Act assumed that an autonomous and fully functional ecosystem  made up of harmonised technical standards, accredited conformity assessment bodies (Notified Bodies) and operational market surveillance authorities in the various countries could be in place within 24 months.

In reality, the process of harmonisation of standards led by the European standardisation organisations (CEN and CENELEC) was not progressing fast enough. The "presumption of conformity" offered by these harmonised standards is key to the concept of compliance to high-risk obligations under the AI Act, with the obligation to comply without the existence of such standards would have brought a significant degree of legal uncertainty and disproportionate costs of compliance, especially for domestic European developers.

In doing so, the EU has tried to keep its digital market competitive. Digital AI Omnibus is not a “deregulatory” approach, but rather a structural adjustment that will not create a “market bottleneck” yet will preserve the risk-based nature of the Act.

The Bifurcated Transparency Regime (2nd of Today)

The split is intentional, and is based on the fact that the rules in the transparency section of the AI Act (Article 50) are less reliant on complex, harmonised rules of third parties and are therefore more technically and legally feasible to implement in short order.

There are four different disclosure regimes in effect from today:

A. Article 50(1) provides for direct interaction. Direct interaction is provided for in Art. 50(1).

Providers or deployers of AI systems that are intended to interact directly with natural persons, such as conversational agents, chatbots, shall clearly and prominently inform of this interaction.

The rule is analyzed from the user autonomy and cognitive manipulation prevention viewpoint. The issue of corporate compliance is whether the requirement to take the steps necessary to comply with the law is obvious from the circumstances. Organisations should carefully review the need for, and the context of, explicit disclosures and consider whether this is necessary or whether passive context will suffice, as regulatory agencies are likely to view the term "obviousness" narrowly.

B. The following articles are available at this address: Synthetic Content and Watermarking (Article 50(2))

Synthetic audio, image, video and text content generated by AI systems should be clearly identified using a machine-readable identifier.[4]

The Digital AI Omnibus has provided a crucial context and time-bounded grace period here; one that is necessarily pragmatic and critical at the same time. New systems on the market after 2 August 2026 will have to meet the requirements immediately, whereas systems already on the market before 2 August 2026 will be given until 2 December 2026 to comply with the requirements.

Moreover, before today, outputs produced using the tools and those manipulated by them are not subject to retroactive marking to avoid retroactive liability and legal uncertainty, since the development of effective, tamper-resistant watermarking is still an engineering challenge.

C. Biometric and Emotion Recognition (Article 50(3))

It is important for people who deploy emotion recognition or biometric categorisation technologies to let people know that they are being exposed to such technologies. [5]

The disclosure requirement directly ties into the General Data Protection Regulation (GDPR), with a specific focus on the legal bases for processing sensitive biometric data. The AI Act's disclosure rule is another formality that adds an extra layer of administrative compliance and also represents an examination of a heightened regulatory focus on AI in the workplace, education, and public space.

D. Deepfakes and Public Interest Media (Article 50(4))

The use of a deepfake or AI-generated text for a public interest topic should be clearly stated by the deployer of the fake text.[6]

The EU's main measure on disinformation. Labeling retroactively is avoided, creating a significant amount of pre-existing synthetic media that remain unlabeled, but also simplifies enforcement of the legislation as it does not need to cover legacy deepfakes created before today.

General-Purpose AI (GPAI): From Guidance to Enforcement

The one-year transitional period for the enforcement of the core obligations under Chapter V of the AI Act has formally ended today; however, the European Commission's AI Office did not have active fining powers in the first 12 months. The core obligations under Chapter V of the AI Act started to apply theoretically today, 2 August 2025, but during the initial 12-month period, the European Commission's AI Office did not have active fining powers. Those enforcement powers are in full effect until today.

In order to cope with this shift, the Commission has been increasingly turning to co-regulatory tools. This strategy is illustrated by the first list of signatories to the voluntary Code of Practice on 30 July 2026.

In the EU context of digital regulation the Code is a voluntary one but when it comes to the standard of care, it actually serves as a “safe harbour”. The general obligations of Article 50 and Chapter V will mean that non-signatories will have to produce independent and robust evidence of compliance, which may lead to direct fine by the regulator in the event of non-compliance.

[1] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) [2024] OJ L 2024/1689.

[2] 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) [2026] OJ L 2026/1744.

[3] ibid.

[4] Regulation (EU) 2024/1689, art 50(2).

[5] Regulation (EU) 2024/1689, art 50(3).

[6] Regulation (EU) 2024/1689, art 50(4).