If your product has a chatbot, AI assistant, or generates AI content for EU users, Article 50 of the EU AI Act applies to you — and the deadline is August 2, 2026. That deadline was not changed by the Digital Omnibus.
Most founders are aware of the larger EU AI Act compliance requirements — the Article 11 technical documentation, the Declaration of Conformity — and have noted with relief that the Digital Omnibus extended those deadlines to December 2027. What many have missed is that Article 50 transparency obligations operate on a separate, earlier timeline that the Omnibus did not touch.
The Digital Omnibus did not move the Article 50(1) deadline. Chatbot disclosure and AI-generated content marking requirements for new systems apply from August 2, 2026 — not December 2026, not December 2027. August 2, 2026.
Article 50 of Regulation (EU) 2024/1689 contains two separate obligations that apply to different types of AI systems.
Any AI system that interacts directly with natural persons — chatbots, virtual assistants, AI customer support tools — must inform those persons that they are interacting with an AI system, unless this is obvious from context. The disclosure must happen at the beginning of the interaction, before the user engages with the system.
The obligation applies when EU users interact with your system, regardless of where your company is incorporated. A US SaaS company whose chatbot handles queries from EU customers is within scope.
There is an exception: if it is "obvious from the context" that the user is interacting with an AI, no additional disclosure is required. In practice, this exception is narrow. A product named "AI Assistant" with a robot icon may qualify. A support chat widget that looks like a human agent conversation almost certainly does not.
AI systems that generate synthetic audio, image, video, or text content must mark that content as AI-generated in a machine-readable format. This is the watermarking obligation.
The Digital Omnibus created a transitional period for existing systems implementing Article 50(2): those systems have until December 2, 2026 to comply. New systems deployed after August 2, 2026 must comply from launch.
| Obligation | Deadline | Notes |
|---|---|---|
| Art. 50(1) — Chatbot / AI disclosure | August 2, 2026 | All systems, new and existing. Not changed by Omnibus. |
| Art. 50(2) — Watermarking, new systems | August 2, 2026 | Systems launched after this date must comply from day one. |
| Art. 50(2) — Watermarking, existing systems | December 2, 2026 | Omnibus transitional period for systems already deployed. |
Article 50 applies to providers of AI systems that interact with users or generate content. In practice, that means:
Notably, Article 50 does not require your system to be high-risk. The transparency obligations apply broadly — a minimal-risk AI chatbot that helps users search a knowledge base must still disclose it is AI. The high-risk classification affects Articles 11, 13, and 47. Article 50 is a separate track.
Not sure if Article 50 applies to your product? Take the free 2-minute risk classification quiz — it includes an Article 50 assessment for chatbot and content generation tools.
There are two components to Article 50 compliance: the technical implementation and the governance documentation.
For Article 50(1), the minimum implementation is a disclosure message shown to users at the start of an AI interaction. The message does not need to be in legal language — it needs to be clear and prominent. Examples that are likely sufficient:
Implementations that are not sufficient:
For Article 50(2), the watermarking requirement is more technically demanding. The European AI Office is still developing technical standards for machine-readable content marking. Practical implementation guidance is expected in 2026. In the interim, a reasonable baseline is: mark AI-generated content visibly (a label or watermark) and implement what machine-readable metadata your toolchain supports (EXIF data, C2PA content credentials where available).
The technical implementation is necessary but not sufficient. National competent authorities enforcing Article 50 may ask for documented evidence of your compliance mechanism — not just a screenshot of your UI. A governance record should show:
This governance record is what regulators will ask for if your product is investigated. A badge in your UI demonstrates good faith; a documented governance record is what closes an enforcement inquiry.
Article 50 breaches fall under Article 99 Tier 2 penalties: up to €15 million or 3% of global annual turnover, whichever is higher. For a startup with €2 million ARR, that is €60,000 of potential exposure — on a product where the compliance record costs $39 to generate.
Enforcement of the EU AI Act is handled by national market surveillance authorities, with the EU AI Office taking a coordination role. Early enforcement activity is expected to focus on the most visible violations — including chatbots that do not disclose they are AI, which are detectable by regulators without any special investigation.
The practical steps are straightforward:
Deadline: August 2, 2026
A formal governance record documenting your Article 50(1) chatbot disclosure and Article 50(2) content marking obligations. Generated from your system details in minutes.
Designed for attorney review and final sign-off. Not legal advice.
Generate My Compliance Record — $39 →One-time fee. Instant PDF delivery. August 2, 2026 deadline.
Possibly not, if it is genuinely "obvious from the context" that users are interacting with an AI under Article 50(1). But "obvious from context" is a legal standard, not a UI judgment call. A clearly branded AI product may qualify for the exception, but documenting your reasoning for relying on it is advisable. If in doubt, add the disclosure — it is two lines of UI copy and substantially reduces your regulatory risk.
The obligation attaches to the provider deploying the chatbot — typically the company integrating it into their product. If you embed a third-party chatbot in your product, check whether the vendor's implementation includes Article 50 disclosure by default. If not, you are responsible for adding it at the integration layer.
Article 50(1) applies to AI systems that "interact directly with natural persons." If your B2B product has end users who interact with an AI chatbot — employees of your business customers, for example — those interactions are within scope. The regulation does not distinguish B2B from B2C at the Article 50 level.
Non-compliance with Article 50 is subject to Article 99 Tier 2 penalties: up to €15 million or 3% of global turnover. National authorities have discretion on enforcement timing and approach, and early enforcement may focus on the most egregious violations. However, non-compliance creates documented regulatory exposure that will not improve over time.
