EU AI Act Deadlines & Fines 2026: August 2 vs. December 2

Last updated: July 21, 2026. More numbers than facts are currently circulating around EU AI Act deadlines and fines: sometimes August 8 is mentioned, sometimes a fine of 35 million euros, sometimes a general postponement to 2027. Most of it is wrong. This article cleanly sorts out the two decisive dates, August 2, 2026 (the core deadline) and December 2, 2026 (the provisional Omnibus postponement), states the correct fine range, and gives you a concrete "by when, what" checklist.

If you publish AI-generated images, videos, or texts, this affects you directly. The good news: the effort is manageable once you know which deadline applies to you and which role you hold.

Key Takeaways

• The core deadline is August 2, 2026, not August 8. From this day, the transparency obligations under Article 50 of the EU AI Act apply.
• The Digital Omnibus (May 2026) postpones only a partial area to December 2, 2026: the machine-readable marking (Art. 50(2)), and only for systems placed on the market before August 2, 2026.
• Visible labeling by deployers (Art. 50(4)) and chatbot disclosure are not postponed.
• The fine is up to 15 million euros or 3% of global annual turnover (Art. 99). The circulating "35 million euros / 7%" figures do NOT apply to labeling.
• Additional risk: competition law cease-and-desist notices, particularly in advertising and e-commerce.
• For D2C, e-commerce, and marketing, the general rule is: you are a deployer and must label visibly.

Contents

1. The one relevant core deadline: August 2, 2026
2. December 2, 2026: what the Digital Omnibus really postpones
3. Deadline table: who must do what by when?
4. Fines: up to 15 million euros, and why the "35 million" figure is a myth
5. Three practical scenarios and their deadlines
6. Checklist: by when must you have done what?
7. Visible label plus metadata in one step
8. Frequently asked questions (FAQ)

The One Relevant Core Deadline: August 2, 2026

The EU AI Act deadlines for labeling have exactly one fixed anchor point. The transparency and labeling obligations under Article 50 of the AI Act apply from August 2, 2026. This date results from the regulation's entry into force on August 1, 2024, plus a 24-month transition period, as also shown by the European Commission's official timeline.

Remember one thing above all: it is not August 8. This misconception haunts numerous articles and social media posts but has no legal basis. It probably arises from confusion with August 1 (entry into force) or a simple digit transposition. Anyone who sets editorial plans, internal deadlines, or compliance roadmaps to August 8 is planning against reality, and gives away six days of buffer.

From August 2, 2026, artificially generated or manipulated content generally must be identifiable as such. The EU AI Act strictly distinguishes between two roles here, and this distinction decides exactly which deadline and which obligation applies to you:

• Provider (Art. 50(2)): whoever develops an AI system or offers it under their own name must mark synthetic image, audio, video, and text outputs in a machine-readable way, "effective, interoperable, robust, and reliable, to the extent technically feasible." Tools for this include IPTC DigitalSourceType/XMP metadata, the C2PA standard (Content Credentials), and watermarks.
• Deployer (Art. 50(4)): whoever uses an AI system to publish content must visibly disclose deepfakes and label AI-generated text on topics of public interest. A lighter standard applies to art and satire.

For the typical D2C, e-commerce, and marketing audience, this means: you are almost always a deployer. You generate an ad image with a generative tool and publish it; your obligation is the visible labeling under paragraph 4. You can read exactly how the two roles are distinguished in our article on the obligations of providers and deployers.

December 2, 2026: What the Digital Omnibus Really Postpones

In May 2026, the European Commission proposed a provisional postponement with the so-called Digital Omnibus, and this is exactly where most misunderstandings arise. What is postponed is not the entire labeling obligation, but only a narrowly defined partial area.

Specifically postponed to December 2, 2026 is exclusively:

• the machine-readable marking under Art. 50(2) (that is, metadata such as IPTC/XMP, C2PA, or watermarks),
• and that only for existing systems, meaning AI systems placed on the market before August 2, 2026.

Everything else stays at August 2:

• The core deadline of 8/2/2026 stands firm.
• Visible labeling by deployers (Art. 50(4)), such as openly labeling deepfakes, is not postponed.
• The chatbot disclosure obligation (users must recognize that they are interacting with an AI) is not postponed.

Why this nuance is so often missed: in many law firm newsletters and press articles, "postponement of the machine-readable marking for legacy systems" gets shortened to "postponement of the labeling obligation." That is grossly misleading. For you as a deployer who continuously produces new AI advertising, the Omnibus changes almost nothing in practice: your visible labeling obligation starts on August 2, 2026.

Important: this postponement is provisional. As the Gibson Dunn analysis of the Omnibus agreement explains, it can still be adjusted during the further legislative process. Relying on it carries a planning risk. The safe strategy is: treat August 2, 2026 as your real deadline and view the December buffer only as a technical grace period for legacy inventory in the machine-readable marking.

Deadline Table: Who Must Do What by When?

The following overview brings role, obligation, and deadline together at a glance:

In short: only the "existing systems" cell slips to December. Everything related to visible labeling is in force on August 2.

Fines for Missing AI Labeling: Up to 15 Million Euros, and Why the "35 Million" Figure Is a Myth

Let's turn to the question that generates the most clicks: how high is the fine for missing AI labeling?

For violations of the transparency obligations under Article 50, the AI Act (fine provisions, Art. 99) sets a range of up to 15 million euros or 3% of global annual turnover, whichever amount is higher. This is the verified, reliable figure. For a corporation with billions in revenue, the percentage value can be well above 15 million euros; for small and medium-sized enterprises, the absolute amount is regularly the upper limit, and the competent authority weighs severity, duration, and intent.

The 35 Million Euro Myth

Time and again, sometimes even on news sites, 35 million euros or 7% is mentioned. For the labeling obligation, this figure is simply wrong. The verified range for violations of the transparency obligations under Article 50 is up to 15 million euros or 3%. The much higher amounts apply to the most serious violations of the AI Act, such as expressly prohibited AI practices like manipulative social scoring, not the transparency and labeling obligations.

In brief:

Anyone who advertises missing AI labeling with the 35 million figure is therefore spreading a factual error. The real range is serious enough; it doesn't need to be dramatized. You can find a detailed breakdown in our article on penalties and cease-and-desist notices for missing AI labeling.

Not Just Authorities: The Cease-and-Desist Risk

Alongside the regulatory fine, there is a second lever that is often faster in practice: competition law. Missing or misleading labeling of AI-generated content, particularly in advertising and e-commerce, can generally be met with a cease-and-desist notice for unfair conduct (under German competition law, the UWG), for example from competitors or associations. Unlike a fine proceeding, which an authority first has to initiate, a cease-and-desist notice can land on your desk within days and bring costs as well as a penalty-backed declaration to cease and desist. Whether a broad wave of cease-and-desist notices actually materializes remains to be seen; you should still factor in the risk of an individual notice. This is particularly relevant in e-commerce and marketing as well as for Meta and Instagram ads. (This is a general risk assessment, not legal advice, see the note at the end.)

Three Practical Scenarios and Their Deadlines

Deadlines become tangible as soon as you apply them to real cases. Three typical constellations:

Scenario 1, the online shop with AI product images. From August 2026, you create AI-generated product or lifestyle images for your shop and your ads. You are a deployer. If the image photorealistically shows real people or scenes (potential deepfake character), the visible labeling obligation applies from August 2, 2026. The Omnibus postponement does not help you here, because it only concerns the machine-readable marking of legacy systems. Practical details on this in the article on AI product images on Amazon and eBay.

Scenario 2, the AI provider with a legacy system. You have been operating a generative image generator since 2025, which you offer to customers. You are a provider, and your system was placed on the market before 8/2/2026. For the machine-readable marking (Art. 50(2)), you provisionally have until December 2, 2026, thanks to the Omnibus. Other obligations, such as a recognizable AI interaction, remain at August 2.

Scenario 3, the partially AI-edited photo. You retouch a real photo with a generative tool (background swapped, objects removed). Whether a labeling obligation exists depends on the extent and context; the EU provides a dedicated "partially AI-modified" icon for this. Details in our article on deepfake labeling.

Checklist: EU AI Act Deadlines & Fines, by When Must You Have Done What?

The following overview brings the EU AI Act deadlines and the fine risk into a practical order.

By August 2, 2026 (fixed, for everyone):

1. Ensure visible labeling of deepfakes and AI-generated image, audio, and video content (deployer obligation, paragraph 4).
2. Design chatbots so that users recognize at the start of the interaction that they are talking to an AI.
3. Label AI texts on topics of public interest, unless editorial control by a human applies.
4. Clarify your role: are you a provider or a deployer? For D2C, e-commerce, and marketing, the general rule is: you are a deployer (paragraph 4).

By December 2, 2026 (provisional, existing systems only):

5. Retrofit machine-readable marking (IPTC/XMP metadata, optionally C2PA/Content Credentials or a watermark) for systems placed on the market before 8/2/2026.

Practical implementation note:

6. Place labeling on the content itself wherever possible, so it is clearly recognizable to users; a blanket note in the legal notice or footer likely won't be sufficient.
7. Use unambiguous wording (e.g., "AI generated"); vague or downplaying terms may contradict the required recognizability.
8. Check the Commission's official, optional EU icons (variants "basic," "fully AI-generated," and "partially AI-modified"; available as SVG and PNG). More on this in the article on EU icons for AI content.
9. Document evidence: who labeled which content and when. A labeling register significantly eases accountability toward authorities.

You can find a detailed foundational breakdown with all roles and exceptions in our pillar article on AI labeling obligations under the EU AI Act and in the compact AI labeling checklist. The authoritative details are in the European Commission's guidelines on the transparency obligations, covered in legal depth in Sidley's compliance analysis.

Generate a Visible Label Plus Metadata in One Step

Depending on your role, labeling covers two levels: the visible label on the content and the machine-readable marking in the metadata. Generating both manually and correctly, including evidence that you labeled on time, is cumbersome without a tool.

That is exactly what our free AI labeling tool is for: you drag the official EU icons onto images and videos, optionally write machine-readable IPTC/XMP metadata (optionally C2PA plus an invisible watermark, see C2PA, IPTC & watermarks in detail), and get evidence with a SHA-256 checksum and timestamp (CSV export) through the integrated labeling register. Batch processing (up to 50 files as a ZIP), a "check file" function, and a REST API are also included. This way, you cover both deadlines, the one on August 2 and the one on December 2, in a single workflow step. You can look up unclear terms in the AI labeling glossary.

Frequently Asked Questions (FAQ)

From when does the AI labeling obligation apply? The labeling obligations under Article 50 of the EU AI Act apply from August 2, 2026. Not August 8, that date is a widespread misconception with no legal basis.

How high is the fine for missing AI labeling? Violations of the transparency obligations (Art. 50) carry a risk of up to 15 million euros or 3% of global annual turnover, whichever amount is higher.

Is the "35 million euro fine" true? No, not for the labeling obligation. For violations of the transparency obligations under Article 50, the verified range is up to 15 million euros or 3%. The circulating 35 million euros applies to the most serious violations of the AI Act, not to labeling.

What was postponed to December 2, 2026 by the Digital Omnibus? Only the machine-readable marking (Art. 50(2)) for systems placed on the market before 8/2/2026 is provisionally postponed. Visible deployer labeling and chatbot disclosure are not postponed.

Is the postponement to December 2 final? No. The Digital Omnibus from May 2026 is provisional and may still change. To be safe, plan with August 2, 2026 as your fixed deadline.

Who must label, the provider or the deployer? Both, but differently. Providers mark content in a machine-readable way (paragraph 2); deployers label content visibly, such as deepfakes (paragraph 4). Anyone publishing AI content in marketing or a shop is usually a deployer.

Can I receive a cease-and-desist notice for missing AI labeling? In principle, yes. Alongside regulatory fines, missing or misleading labeling, especially in advertising and e-commerce, can become relevant under competition law (the UWG). This is a general risk assessment, not a substitute for legal advice.

Is a note in the legal notice or footer enough? Probably not. Labeling should happen on the content itself, so it is clearly recognizable to users.

Are the EU icons mandatory? No. The official EU icons are optional. They make clear, consistent labeling easier but are not required. What matters is that the AI character is recognizable.

Does the obligation also apply to AI-generated texts? Yes, but with limits: AI texts on topics of public interest must be labeled, unless human editorial control applies. Details in the article on labeling ChatGPT texts.

Sources

• European Commission, timeline for implementing the EU AI Act: https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act
• European Commission, guidelines on the transparency obligations (Art. 50): https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems
• European Commission, official EU icons for labeling AI-generated content: https://digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content
• Article 50 of the AI Act (full text): https://artificialintelligenceact.eu/article/50/
• Sidley, EU AI Act Transparency Obligations (compliance by August 2, 2026): https://datamatters.sidley.com/2026/06/24/eu-ai-act-transparency-obligations-preparing-for-compliance-by-2-august-2026/
• Gibson Dunn, EU AI Act Omnibus Agreement (postponements and changes): https://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes/

This article reflects the status as of July 21, 2026 and does not replace legal advice. The Omnibus postponement is provisional and may change. For your specific case, please consult qualified legal advice.