The AI Act deadline the Omnibus did not move

On Sunday, the European Commission's AI Office starts enforcing the parts of the AI Act that govern general-purpose AI models and the transparency rules in Article 50. According to the Commission's press release of 31 July, chatbots must tell users they are AI, deepfakes must be labelled, and AI-generated content must carry machine-readable marks.¹ A lot of companies deploying AI in Europe appear to believe this deadline moved. It did not.

I have been reading compliance coverage of the EU's Digital Omnibus for the past two months, and what strikes me is how much of it collapsed a specific extension into a general one. The Omnibus, in force from 27 July, did push back the timelines for high-risk AI systems under Annex III, according to Politico and Euractiv.² It did not push back GPAI enforcement or Article 50. The Commission's own press release, three days before enforcement begins, is in effect a correction to the story a lot of deployers were telling themselves.

Who this actually lands on. The frontier labs are ready. OpenAI, Google DeepMind, Anthropic, Meta and Mistral are, according to Reuters and the Compliance Hub Wiki synthesis, signatories to the GPAI Code of Practice, and the Commission has published a first list of more than 180 organisations that have signed the transparency Code that operationalises Article 50.¹³⁴ These are the companies with Brussels offices, standing legal teams, and the ability to track a legislative negotiation in real time as it collapsed in April, revived in modified form in May, and settled into the Omnibus in July.³

The companies I am thinking about are the ones further down the stack. The mid-sized SaaS firm embedding a third-party model into a customer-support flow. The healthcare start-up whose triage chatbot talks to patients in three EU languages. The public-sector contractor whose citizen-facing assistant answers benefits questions. Under Article 50, these are deployers, and deployers carry the chatbot-disclosure and deepfake-labelling duties directly, according to the Fieldfisher and AI Policy Discussions readings of the text.⁵ They do not get to point upstream at the model provider.

What the rule actually asks for. Article 50, as summarised by the Commission and by legal analyses of the text, breaks into a few obligations that hit different actors.¹⁵ Providers of systems generating synthetic audio, image, video or text must mark the output in a machine-readable format detectable as AI-generated. Deployers of interactive AI systems must inform users they are dealing with AI, unless it is obvious from context. Deployers of deepfake systems must disclose that the content is artificially created. Deployers using AI to generate text on matters of public interest must disclose AI authorship, with an exemption for minor editing.

The watermarking piece is the one carve-out worth being precise about. According to a Latham & Watkins analysis of the Omnibus text, the deferral of the Article 50(2) machine-readable-mark obligation to 2 December 2026 applies to generative systems already placed on the market before 2 August — a grandfathering rule for pre-existing systems, not a general text-only exemption.⁶ Systems launched from Sunday onward carry the obligation from Sunday.

The distributional shape of the compliance gap. I want to take seriously the argument that a lot of European AI deployment is now in the hands of firms sophisticated enough to have tracked the Omnibus negotiations. That is true of the top of the market. It is less true the further down you go, and the pattern of the last two months has been that trade coverage compressed "some AI Act deadlines extended" into "the AI Act deadline extended". The Compliance Hub Wiki synthesis puts it plainly: a lot of compliance teams missed the distinction.³

Who benefits from that confusion? Not, obviously, the deployers who slowed down and now find themselves out of time. Not the users the transparency rules exist to protect. The firms that kept building compliance infrastructure through the noise, mostly the largest ones, arrive at Sunday with a working machine, and the ability to point at their Code of Practice signature. The firms that read the Omnibus headlines and paused arrive without one. The rule is universal; the readiness is not.

Fines and enforcement, in the ranges the coverage cites. The Commission's press release does not itself state the fine ceiling. Legal analyses in this space, including Fieldfisher and AI Policy Discussions, cite maximum fines under the AI Act for Article 50 and GPAI violations of up to fifteen million euros or three percent of global annual turnover, whichever is higher.⁵ Reuters reports that the AI Office has established complaint channels, a whistleblower tool, and a downstream-provider complaints channel, and that officials describe enforcement as immediate rather than phased.⁴ IAPP's analysis notes that Article 50 applies to any provider or deployer whose output is received by users in the EU, regardless of where the provider is incorporated, on logic that mirrors the GDPR's territorial reach.⁷

That last point is why the American frontier labs are inside this, not outside it. Their EU-facing products fall under Article 50 disclosure from Sunday.

What is genuinely open. The AI Office has the legal authority from day one. What it does not yet have is case law. "Obvious from the circumstances and context of use" is the exemption a lot of deployers will lean on for the chatbot-disclosure obligation, and legal commentators note it has not been tested by enforcement decisions.⁵ The companies and civil-society organisations that file the first complaints will help shape what counts as obvious, and what does not. That is a real prize, and it is the kind of prize that tends to go to whoever has the standing and the resources to be first through the door.

The Commission publishes its Article 50 enforcement approach in a set of guidelines that the coverage I have read describes but that I have not seen in a fetched primary version. The AI Office's first enforcement actions, and the first national-authority decisions on deployer obligations, are what I will be watching for. Sunday is not the story ending; it is the day the clock on those cases starts.

Glossary

GPAI General-purpose AI model; a foundation model that can be adapted to many downstream tasks.

Article 50 The AI Act provision covering transparency obligations for interactive AI systems, deepfakes, and AI-generated content.

AI Office The unit inside the European Commission with enforcement authority over GPAI models under the AI Act.

Digital Omnibus The 2026 EU regulation that adjusted several AI Act compliance deadlines, in force from 27 July 2026.

Code of Practice A voluntary industry code operationalising specific AI Act obligations; signature signals commitment to the code's approach.

Annex III The AI Act annex listing high-risk AI use cases; systems in these categories carry the heaviest compliance load.

---

Footnotes

Footnotes

1. European Commission, "Commission starts enforcing AI Act rules and new transparency requirements on 2 August," digital-strategy.ec.europa.eu, 31 July 2026. https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august ↩ ↩² ↩³

2. Politico EU, "EU AI Act high-risk deadline pushed back in Digital Omnibus deal," politico.eu, 2026. https://www.politico.eu/article/eu-ai-act-omnibus-amendment-high-risk-deadline-extension-2026/ ↩

3. Compliance Hub Wiki, "60 Days to EU AI Act Enforcement," compliancehub.wiki, 29 May 2026. https://compliancehub.wiki/eu-ai-act-august-2-2026-60-day-countdown-synthesis ↩ ↩² ↩³

4. Reuters, "EU AI Office ready to enforce AI Act GPAI rules from August," reuters.com, 30 July 2026. https://www.reuters.com/technology/eu-ai-office-enforcement-2026-07-30/ ↩ ↩²

5. AI Policy Discussions, "EU's AI Act Article 50 Transparency Obligations," aipolicydiscussions.com, 22 October 2025. https://aipolicydiscussions.com/2025/10/22/eus-ai-act-article-50-transparency-obligations/ ↩ ↩² ↩³ ↩⁴

6. Latham & Watkins, "AI Act Update: EU Resolves to Change Rules and Extend Deadlines," lw.com. https://www.lw.com/en/insights/ai-act-update-eu-resolves-to-change-rules-and-extend-deadlines ↩

7. IAPP, "EU AI Act Article 50 — extraterritorial scope explained," iapp.org, 2026. https://iapp.org/news/a/eu-ai-act-article-50-extraterritorial-scope/ ↩