What Applies Under the AI Act From 2 August 2026

Last updated: 22 July 2026

Legal status. The Digital Omnibus on AI, which rewrites the AI Act’s remaining deadlines, was signed on 8 July 2026 but had not been published in the Official Journal as of 22 July 2026. The European Parliament’s Legislative Observatory lists procedure 2025/0359(COD) as completed and awaiting publication, after Parliament adopted the text on 16 June 2026 by 423 votes to 57 with 174 abstentions and the Council followed on 29 June. The amending regulation enters into force on the third day after publication. Until then, the original timeline in Regulation (EU) 2024/1689 remains the binding law. We will update this article when the OJ reference exists.

The AI Act 2 August 2026 deadline still arrives, but it now carries far less than the original timeline built into two years of compliance calendars. Assuming the Digital Omnibus is published in time, what switches on is the Article 50 transparency regime and the Commission’s power to fine GPAI model providers. The high-risk obligations move to 2 December 2027 and 2 August 2028. For a company with 50 to 1,000 employees, the pre-August work is narrow, and most of it depends on which side of the provider and deployer line each duty sits.

Provision	Article	Who owes it	Status on 2 August 2026
AI interaction disclosure (chatbots, voice assistants)	Art 50(1)	Provider	Applies
Machine-readable marking of synthetic content, systems placed on the market from 2 August 2026	Art 50(2)	Provider	Applies
Emotion recognition and biometric categorisation disclosure	Art 50(3)	Deployer	Applies
Deepfake and public-interest text labelling	Art 50(4)	Deployer	Applies
Commission fining power over GPAI model providers	Art 101	GPAI model provider	Applies
National market surveillance of provisions applicable by this date, including AI literacy	Ch IX, Art 74; Art 4	All operators	Begins
Art 50(2) marking for generative systems placed on the market before 2 August 2026	Art 50(2)	Provider	2 December 2026
New prohibitions on AI-generated non-consensual intimate imagery and CSAM	Art 5 (amended)	All operators	2 December 2026
National regulatory sandboxes	Art 57	Member States	Deferred to 2 August 2027
Compliance date for GPAI models placed on the market before 2 August 2025	Art 111(3)	GPAI model provider	2 August 2027, original AI Act timeline, unaffected by the Omnibus
Stand-alone high-risk obligations (Annex III)	Ch III	Providers and deployers	Deferred to 2 December 2027
Embedded high-risk obligations (Annex I products)	Ch III	Providers and deployers	Deferred to 2 August 2028

Every deferral in this table depends on the Omnibus reaching the Official Journal, with one exception. The 2 August 2027 date for legacy GPAI models sits in Article 111(3) of the original Act and holds either way.

What is the August 2 AI Act deadline?

2 August 2026 is the general application date of the EU AI Act under Article 113, the day the regulation’s remaining provisions were originally due to apply across the single market. After the Digital Omnibus, the date delivers the Article 50 transparency duties and the Article 101 GPAI fining power, not the high-risk regime.

The Union closed the legislative procedure to move 2 August 2026 twenty-five days before 2 August 2026, and if the Official Journal misses the window, the old dates bite in full. Our read is that publication lands in July, but a prediction is not a legal basis, which is why the table above marks every deferral as conditional. It also settles a claim in circulation since May, that the delayed dates are already binding law. As of 22 July 2026 they are not, and any guidance asserting either the old position or the new one without a publication caveat is describing a law that does not quite exist yet.

The AI Act 2 August 2026 Duties Split Evenly Between Providers and Deployers

Article 50 contains four transparency duties, and the Commission’s final guidelines, published on 20 July 2026, confirm the allocation. Articles 50(1) and 50(2) bind providers. Articles 50(3) and 50(4) bind deployers.

On 2 August that split decides what is your engineering and what is your supplier’s contract. For the systems you licence, the chatbot disclosure and the machine-readable marking travel with the provider; the emotion recognition and deepfake disclosures stay with you and no procurement clause moves them. The full breakdown of all four duties, the exemptions for edited public-interest text and assistive editing, and the marking question for retrained models sit in our AI Act transparency requirements guide. One date-specific detail belongs here. Providers of generative systems already on the market on 2 August get the Article 50(2) marking duty from 2 December 2026, and recital 38 of the signed Omnibus describes that verbatim as a transitional period of four months, which settles the three-month figure in the Council’s press release against the calendar.

The Approved Compliance Route Opened in June and Signing Stays Open Past 2 August

There is an officially endorsed way to demonstrate Article 50 compliance, and it changes the task list. The Code of Practice on Transparency of AI-Generated Content, published on 10 June 2026, received the Commission’s adequacy opinion on 8 July 2026 and the AI Board’s assessment the following day, covering Articles 50(2), (4) and (5). Signatories can rely on adherence to the code to demonstrate compliance, with supervision focused on that adherence, though the Commission is explicit that adherence is not conclusive evidence. Non-signatories must demonstrate compliance through other means, assessed by their market surveillance authority. Per the Commission’s signing FAQ, the code has separate provider and deployer sections that can be signed independently. The deadline for the first public signatory list passed at 18.00 CET on 22 July 2026, but signing remains open and the AI Office encourages submissions until 27 July 2026. In our assessment, a deployer with material 50(4) exposure should treat the signing decision as part of the August preparation, not an afterthought, because the alternative is proving compliance alone to a national authority with a fresh mandate.

Article 101 Is the Only Fine Power That Newly Arrives on 2 August 2026

The one enforcement mechanism that activates on 2 August 2026 is the Commission’s power under Article 101 to fine providers of general-purpose AI models up to 3% of worldwide annual turnover or EUR 15 million, whichever is higher. Every other penalty power in Chapter XII has been available to national authorities since 2 August 2025.

Two clarifications de-escalate most of the fear attached to the date. GPAI obligations themselves are not new; Chapter V has applied since 2 August 2025, so what arrives for model providers is Commission enforcement, not new duties. And the EUR 35 million and 7% ceiling circulating in coverage of this date belongs to Article 99(3) and prohibited practices, a tier that reaches deployers too but has nothing to do with GPAI or with 2 August. A company with 50 to 1,000 employees is almost never a GPAI provider, with one caveat worth checking. Place a system on the market under your own name or trademark, modify it substantially, or repurpose it, and you can cross from deployer to provider. Our AI Act Risk Classifier walks through where a given system sits, and the tier-by-tier fines breakdown, including the SME cap that inverts the ceilings, is in the transparency guide linked above.

The Deferred Dates Are a Planning Baseline Until Publication, and a Warning After a Miss

The forward calendar in the table binds only once the Omnibus enters into force, and the fixed dates replaced the Commission’s original standards-linked trigger, which in our assessment is the most useful thing the Omnibus did. If publication slips past 2 August, the original law applies in the gap, including the Annex III high-risk regime, so a company with candidate high-risk systems should know each system’s classification status now rather than bet on the Official Journal’s print schedule. ⚠️ One open point survives into the final text. Article 111(2) exempts legacy high-risk systems unless they undergo significant changes in their designs, while Article 3(23) defines the separate term substantial modification, and nothing in the text connects the two, so the boundary of the legacy exemption is a question the text does not answer. All confirmed dates across this and the other seven frameworks sit in our EU Compliance Deadline Tracker, updated the day the OJ reference appears.

The Scope Inventory and the Code Decision Are Work You Can Finish Before August

For deployers, which is nearly everyone reading this, three tasks fit into the remaining days. Inventory every AI system in use and mark each against the four Article 50 duties and the provider and deployer split, since that split determines whether the fix is your engineering or your supplier’s contract. Decide the Code of Practice question for anything with 50(4) exposure. Document your Article 4 AI literacy measures, because national enforcement of that duty also begins on 2 August 2026 even though the Omnibus softens the wording from ensuring literacy to supporting it. If you use emotion recognition anywhere, the Article 5 screen comes before everything else. And the high-risk programme now points at 2 December 2027, with the confirmed dates held in our EU Compliance Deadline Tracker as each one hardens.