EU AI Act Article 50 for UK businesses — what went live in August 2026 and what lands on 2 December

Time-sensitive (checked against the European Commission timeline on 1 October 2026): Article 50 transparency rules apply from 2 August 2026. The Commission lists 2 December 2026 for the Article 50(2) transition for certain generative systems already on the market before 2 August and for new prohibitions covering systems that generate non-consensual sexual deepfakes or child sexual abuse material. The same timeline lists high-risk Annex III application from 2 December 2027 and Annex I product rules from 2 August 2028. This is a planning summary, not legal advice; confirm the current AI Act Service Desk timeline, legislation and qualified counsel before acting.

On 2 August 2026, EU Article 50 transparency obligations (and related supervisory powers) took effect for in-scope systems. Separately, high-risk application timelines were reshaped under the AI Act Omnibus, which entered into force on 27 July 2026 — so the viral slogan “everything high-risk was due on 2 August 2026” is a bad UK SME brief.

This is a buying and roadmap note for UK teams that ship chatbots, copilots, or synthetic media into the EU or under EU-regulated contracts. It is not legal advice. If you process only UK-domestic data with no EU nexus, you still have UK duties (ICO guidance and related statutes), but Art.50 is not your primary calendar. If you sell into the EU, ignore Art.50 at your peril.

What went live versus what was deferred

Article 50 is about transparency: making it clear to people when they are interacting with AI, and handling certain synthetic content disclosures. That is different from the full high-risk conformity assessment machinery that social posts often mash into one “AI Act day.” The Omnibus deferred and reshaped parts of the high-risk runway — track those dates on a legal timeline, not from a LinkedIn carousel.

Article 50 also covers machine-readable marking of AI-generated content (Art. 50(2)) and deepfake / public-interest text disclosure (Art. 50(4)–(5)). The Commission’s Code of Practice on Transparency of AI-generated Content is the voluntary tool providers and deployers can sign to demonstrate compliance with those marking and labelling obligations. The Code does not change the underlying 2 August 2026 effective date. Treat the technical marking work as a finish-the-engineering item, not as a reason to delay basic “you are chatting with AI” labels in the product UI.

The 2 December 2026 date, and what it does not mean

The Commission's AI Act Service Desk timeline lists 2 December 2026 for two changes:

• The Article 50(2) transition ends. Certain providers of AI systems that generate synthetic content, including general-purpose AI systems, get until 2 December 2026 to meet the machine-readable marking duty. This applies only if the system was placed on the market before 2 August 2026. Morrison Foerster describes it as a grandfathering delay for existing systems.
• New prohibitions apply. AI systems that generate non-consensual sexual deepfakes or child sexual abuse material become prohibited practices.

The transition is narrow. It covers providers of existing generative systems, and only for the marking duty in Art. 50(2). It is not a general grace period. It does not delay the Art. 50(1) duty to tell people they are interacting with AI, or the deployer disclosure duties in Art. 50(4). A system you launched after 2 August 2026 never had the transition. If you are a deployer rather than a provider, your labelling duties already apply. So the advice does not change: label now. If you are a provider of an older system, use the time to finish the marking pipeline, not to postpone it.

What the Commission published since August

• EU icons for labelling AI content. The Commission has published a set of icons that deployers can use to label deepfakes and AI-generated public-interest text under Art. 50(4). There is a basic icon, a "fully AI-generated" icon and a "partially AI-modified" icon, each in four colour variants. Using them is optional. The labelling duty is not. User testing found the icons worked better with a short text label beside them.
• Article 50 guidance and a voluntary Code. The Commission's Code of Practice supports compliance with Article 50 marking and labelling duties. It is voluntary; the underlying transparency requirements are legal obligations. The Commission says the code complements its Article 50 guidelines.
• Code of Practice sign-ups. As of 9 September 2026, 82 providers and 152 deployers had signed the Code of Practice on Transparency of AI-generated Content. Morrison Foerster notes that the Commission's Q&A expects authorities to ask non-signatories for more detailed information.

If your team uses an AI writing tool for customer-facing or public-interest content, check what it marks and what you still have to label. The AI Writing Tool Finder is a starting point. Human review with named editorial responsibility is the exemption for AI-assisted text, so write that process down.

What UK buyers should do this month

1. Inventory customer-facing chatbots, support agents, marketing generators, and synthetic media tools that touch EU users or EU contracts.
2. Label AI interaction and synthetic content where transparency rules apply — in the product UI and in customer-facing copy, not only in a privacy policy footnote.
3. Split calendars: transparency live dates, machine-readable marking work, and high-risk conformity workstreams get separate owners.
4. Vendor diligence: ask SaaS vendors (help desk AI, writing tools, meeting notetakers) how they handle Art.50-style transparency and what you must surface in your own UI.
5. Parallel UK regime: ICO / UK AI governance expectations still apply to UK processing. EU Art.50 compliance is not a substitute for UK advice.

For model and API cliffs on nearby calendars (Sonnet pricing, Copilot promo pools, model EOLs), keep the AI release watchlist open beside the legal tracker so engineering and counsel share one September view.

Who is most likely in scope for a UK SME

• B2B SaaS with EU customers using an in-product AI assistant.
• Agencies generating synthetic images or video for EU brands without clear disclosure.
• Support teams rolling out resolution agents that chat as if human by default.
• HR or recruiting tools that interview or score candidates with generative models for EU roles.

Who may be less immediately pressured: purely internal tools with no EU user-facing interaction and clear contractual containment — still verify with counsel; “internal” is not a magic exemption when outputs reach customers.

Common mistakes

• Treating 2 August as a universal high-risk compliance cliff for every AI feature.
• Reading the 2 December 2026 Art. 50(2) transition as a general grace period. It only covers marking by providers of systems already on the market before 2 August 2026.
• Assuming a US vendor’s ToS automatically covers your EU customer disclosures.
• Letting marketing run synthetic campaigns while product and legal have no shared inventory.
• Ignoring UK duties because an EU checklist feels complete.

A practical two-week plan

Week one: build the inventory spreadsheet (system, audience geography, disclosure today, owner). Week two: ship minimum viable transparency labels on the top three customer-facing surfaces, open vendor tickets, and book counsel for high-risk classification where stakes are real. Engineering can continue marking-pipeline work in parallel, but basic user-facing disclosure should not wait on it.

Bottom line

Article 50 transparency is live as of 2 August 2026 for in-scope systems. The 2 December 2026 date ends a narrow marking transition for existing generative systems and adds new deepfake prohibitions. High-risk timelines are a different story under the Omnibus (Annex III → 2 December 2027; Annex I → 2 August 2028). UK SMEs should inventory, label, track machine-readable marking work separately, and keep UK duties in parallel. Pair this note with the watchlist for commercial AI cliffs so “compliance week” and “pricing week” do not collide by accident. Get qualified legal review before you call anything “compliant.”

Sources

• European Commission AI Act Service Desk, Timeline for the implementation of the EU AI Act (incorporates Digital Omnibus amendments; dates for Article 50, the transition, new prohibitions, and high-risk systems). https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act — accessed 2026-10-01.
• European Commission, AI Omnibus enters into force (27 July 2026; amended Annex III and Annex I application dates; new prohibitions). https://digital-strategy.ec.europa.eu/en/news/ai-omnibus-enters-force — accessed 2026-10-01.
• European Commission, Commission starts enforcing AI Act rules and new transparency requirements on 2 August (Article 50 transparency obligations and enforcement start). https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august — accessed 2026-10-01.
• European Commission, Code of Practice on Transparency of AI-generated Content (voluntary code supporting legally binding Article 50 obligations; code complements Commission guidelines). https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content — accessed 2026-10-01.
• European Commission, EU icons for labelling AI-generated content. https://digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content — accessed 2026-09-26.
• Morrison Foerster, Full Transparency: Organizations Sign EU Code (11 Sep 2026; 82 providers and 152 deployers as of 9 Sep; final Art. 50 Guidelines). https://www.mofo.com/resources/insights/260911-full-transparency-organizations-sign-eu-code — accessed 2026-09-26.
• European Commission, AI Act (Shaping Europe’s digital future) — last updated 3 August 2026. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai — accessed 2026-08-28.
• European Commission, AI Omnibus enters into force — 27 July 2026. https://digital-strategy.ec.europa.eu/en/news/ai-omnibus-enters-force — accessed 2026-08-28.
• European Commission, Code of Practice on Transparency of AI-generated Content — last updated 31 July 2026. https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content — accessed 2026-08-28.
• European Commission press release, Commission starts enforcing AI Act rules and new transparency requirements on 2 August — 31 July 2026. https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august — accessed 2026-08-28.
• Future of Life Institute, Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems (Date of entry into force: 2 August 2026). https://artificialintelligenceact.eu/article/50/ — accessed 2026-08-28.
• ICO, Artificial intelligence guidance landing page. https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/ — accessed 2026-08-28.