Have an AI feature on your site? What the EU AI Act asks of it now, and what’s due by 2 December 2026

Key points

• The Article 50 transparency duties of the EU AI Act have applied since 2 August 2026.
• Only the machine-readable marking duty got more time, and only for generative AI systems placed on the market before 2 August 2026: their providers have until 2 December 2026.
• The AI notice in a chatbot and the visible labels on deepfakes and public-interest text did not get extra time.
• Where your company is based does not decide it: the Act reaches providers and deployers outside the EU when their AI system’s output is used in the EU.
• High-risk rules are not due yet: 2 December 2027 for Annex III uses such as hiring and credit, and 2 August 2028 for AI in regulated products.

What changed, and when

Article 50 of the AI Act, Regulation (EU) 2024/1689, sets the transparency duties for chatbots and generated content. They have applied since 2 August 2026. In July 2026 the EU amended the Act with Regulation (EU) 2026/1744, the Digital Omnibus on AI, published in the Official Journal on 24 July 2026 and in force since 27 July 2026. It added one transitional period to Article 50 and moved the high-risk rules later.

The 2 December 2026 date is narrow. The amended Article 111 gives providers of AI systems that generate synthetic audio, images, video or text, and that were placed on the market before 2 August 2026, until then to comply with Article 50(2), the marking duty. The recitals call it a transitional period of four months. It does not cover the AI notice, and it does not cover the deployer’s visible labels. The Commission also says content generated before 2 August 2026 does not have to be labeled retroactively, though it encourages it.

Our guide Does the EU AI Act apply to my AI app or chatbot? covers the risk tiers, the roles and the penalties in full. This post is the short version of what is due now.

Who this reaches

Article 2 applies the Act to providers that place AI systems on the EU market wherever they are established, and to providers and deployers in other countries when the output of their AI system is used in the EU. A company in India, the US or Brazil whose chatbot answers people in the EU is in scope for those people.

• If you offer a chatbot or AI feature under your own name, even one built on another company’s model, you are generally its provider. The AI notice and the machine-readable marking are your duties, though the marking may come partly from your model vendor.
• If you use an AI tool to produce what you publish, such as images, video or articles, you are its deployer. The visible labels on deepfakes and on public-interest text are your duties.
• Many startups are both: they run their own AI feature and publish with AI tools.

A checklist before 2 December 2026

For a website or app with a chatbot, AI-written content or generated images:

1. List every AI feature and what it produces: chat replies, written text, images, audio or video. Note for each whether you are its provider, its deployer or both, and which model or vendor sits underneath it.
2. Note when each feature first went live. If it was on the market before 2 August 2026, the marking deadline is 2 December 2026. If it launched later, marking already applies.
3. Check the AI notice now; it is not part of the December date. The first message of a chat or voice conversation should say plainly that the person is talking to an AI, in a way a screen reader also announces.
4. Ask your model vendor in writing how the content it generates is marked in a machine-readable way and how that mark can be detected. Then check that your own processing, such as resizing, compressing or stripping file data, does not remove it.
5. Know what falls outside the marking duty: assistive editing that does not substantially change the input, and, according to the Commission, outputs such as source code and short sequences of numbers, symbols or letters.
6. Add visible labels where you publish deepfakes, or AI-written text that informs the public on matters of public interest and has had no human review under someone’s editorial responsibility. A machine-readable mark alone does not meet this duty.
7. If you label, you may use the Commission’s free EU icons for AI-generated content. They are optional and do not prove compliance on their own; the Commission asks that a label be visible at first exposure, not hidden behind overlays, and readable by assistive technology.
8. Write down what you checked and when: each feature, its model, how its output is marked, where labels appear. That record answers most questions a customer or an authority will ask.

The Commission’s Code of Practice on Transparency of AI-generated Content, published on 10 June 2026, sets out practical measures for both roles. Signing it is voluntary; the Commission says about 190 organizations had signed by the end of July 2026. Those who do not sign must show they meet the duties by other adequate means. For the product side of a chatbot, see Adding an AI chatbot to your website.

What is not due yet

• High-risk rules for Annex III uses, such as hiring, education, credit scoring and access to essential services: 2 December 2027, moved from 2 August 2026 by the Omnibus.
• High-risk rules for AI built into products covered by EU product laws, such as machinery, toys or medical devices: 2 August 2028, moved from 2 August 2027.

Not due yet is not the same as nothing to do. If your AI helps decide who gets hired, admitted, insured or given credit, those rules will apply to it, and the logging, human oversight and documentation they ask for take time to build. The controls in How to stop an AI agent from doing damage are a start. The earlier duties have not moved: the first eight bans have applied since 2 February 2025, and the rules for general-purpose AI models since 2 August 2025.

What this post is, and is not

This post is general information, written on 3 November 2026 from the official texts and Commission pages listed below. This is not legal advice. For a decision about your product, talk to a lawyer who works with EU technology law, and check the Commission’s AI Act pages again before 2 December, because the guidance was still being added to in October 2026.

When BYD Tech builds an AI feature through AI agent development, the code lives in the client’s GitHub from day one, IP is assigned to the client, the founders are on the calls, and AI-assisted work is reviewed by an engineer.