Who this covers
Founders and operators of a SaaS product with an AI assistant or generative output that people in the EU can reach. A US, Canadian, or UK headquarters does not take you out. Article 2 of the AI Act covers providers who place a system on the Union market wherever they sit, and providers in a third country whose output is used in the Union. Unsure plus EU users: assume Article 2(1)(a) or (c).
The FAQ tests for a chatbot notice
The Commission’s Article 50 FAQ (updated 24 July 2026) and the 20 July 2026 guidelines set out when Article 50(1) applies. Providers must design the system so people are informed they are interacting with AI, unless that is obvious to a reasonably well-informed person in that context. The FAQ’s tests:
- It is an AI system.
- It is built for a genuine two-way exchange.
- The AI itself talks to the person.
- The other party is a natural person.
Background or machine-to-machine systems are out. The “obvious” exception is to be read narrowly. A product named “AI Copilot” is not automatically obvious if a careful person could think they reached support staff.
What to ship
Notice must land at the start of the first interaction. Put it in the product, not in a privacy policy. Visible without a click. A footer or a line in the Terms will not match “first interaction, clear and distinguishable.”
This is an AI assistant. You are not speaking with a person.
That is the first-message or persistent-chrome line from the sample brief. Keep a dated screenshot and the commit hash.
What the Digital Omnibus actually moved
The AI Act entered into force on 1 August 2024. The general application date was always 2 August 2026. The Digital Omnibus on AI, Regulation (EU) 2026/1744, was adopted on 8 July 2026, published on 24 July, and entered into force on 27 July. It keeps the general application date at 2 August 2026.
Recital 40 moves Annex III stand-alone high-risk use cases (employment, education, credit, essential services, certain biometrics) to 2 December 2027, and high-risk AI inside Annex I products to 2 August 2028. Recital 38 is the only Article 50 concession: four extra months, only for providers who had already placed a generative system on the market before 2 August, and only for the marking duty in Article 50(2). That date is 2 December 2026. Later launches do not get that grace period. Content generated before 2 August does not have to be labelled retroactively.
If your team saw “high-risk delayed to 2027” and closed the ticket, they read the wrong article. Chatbot notice and deepfake labels are already due. Marks on systems launched on or after 2 August are already due. Hiring tools, credit scoring, and education scoring are the high-risk list — not a support bot.
Other live Article 50 duties
This week: the Article 50(1) notice, plus any deployer labels you already owe.
- Tell people exposed to emotion recognition or biometric categorisation that the system is running (Article 50(3)).
- Label deepfakes on first exposure with a label a person can see or hear (Article 50(4)). A hidden watermark from the model vendor is not enough. Artistic or satirical work still needs a disclosure.
- Label AI text published to inform the public on a public-interest topic unless someone with judgement reviewed the substance and an editor takes responsibility. Spell-check is not review.
- Providers must also mark synthetic audio, image, video, or text so it can be detected as AI-generated (Article 50(2)), with carve-outs for short codes, source code, output that never reaches a person, and assistive standard editing that does not substantially change the input.
Do not assume a vendor’s mark covers your UI. Wrapping a frontier model does not label your first message.
Enforcement and fines
From 2 August the AI Office and national authorities have their powers. National market surveillance authorities handle most Article 50 cases. The AI Office handles GPAI models, systems from the same provider or group as the underlying model, and systems inside designated very large platforms or search engines. A public complaints tool is live.
Article 50-type fines can reach €15 million or 3% of worldwide turnover, whichever is higher. As of 27 August 2026 there was no official announcement of a first Article 50 fine or a named investigation against a specific startup.
More than 180 organisations had signed the Code of Practice on Transparency of AI-generated Content by the 31 July press note (the code page says about 190 by late July). On 8 July the Commission found the code adequate for Articles 50(2), (4), and (5). Signing is voluntary and is not conclusive proof of compliance. It also does not operationalise the Article 50(1) chatbot notice.
This week
- Inventory every EU-reachable surface that talks back, generates media, scores a person, or infers emotion or a biometric category. Mark provider vs deployer; shipped before or after 2 August; likely 50(1), 50(2), 50(3), 50(4), or Annex III.
- Ship the 50(1) notice in the first message or persistent chrome.
- Label AI image, audio, or video that could pass as real, and unreviewed public-interest AI text.
- Open a 2 December ticket for machine-readable marks if the generative system was on the market before 2 August 2026.
- Skip the high-risk build unless the inventory says Annex III. That is a 2027 programme and a lawyer, not this week’s ship.
The cheap move is a notice and a screenshot. The expensive move is explaining why you thought 2027 applied to a chatbot.
Sources
Operator memo from official EU pages. Full recommendation, risks, and this-week list: Ship the chatbot notice. The Omnibus delay does not cover Article 50.
- European Commission, enforcement framework of the AI Act, updated 24 August 2026. digital-strategy.ec.europa.eu/en/policies/enforcement-ai-act
- European Commission, enforcement starts 2 August, 31 July 2026. Commission starts enforcing AI Act rules
- European Commission, Article 50 FAQ, updated 24 July 2026. Transparency obligations under Article 50
- European Commission, Article 50 guidelines, 20 July 2026. Guidelines on transparency obligations
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), OJ L 2026/1744, 24 July 2026. EUR-Lex 32026R1744
- AI Act Service Desk, Article 2 (scope). Article 2