EU AI Act Article 50(1): adding a chatbot AI-disclosure notice in 20 lines

Article 50 of the EU AI Act (Regulation (EU) 2024/1689) has applied since 2 August 2026. If your product lets people chat with an AI system and those people are in the EU, Article 50(1) is the rule that usually shows up first in an engineering backlog.

This post is for developers. It is readiness support, not legal advice. This article was AI-assisted and then edited.

#

What Article 50(1) actually says In plain terms, Article 50(1) requires that providers of AI systems intended to interact directly with natural persons design those systems so that the person is informed that they are interacting with an AI system, unless that is obvious from the circumstances and the context of use.

Official text: Regulation (EU) 2024/1689, Article 50. The European Commission has also published transparency guidance and FAQ material for these duties. Start here: Commission enforcement and transparency requirements from 2 August 2026.

Important related deadlines many U.S. teams mix up:

Article 50 as a whole has applied since2 August 2026 (see Art. 113 of Regulation 2024/1689).

• A narrow grace for themachine-readable marking duty in Article 50(2) for some generative systems placed on the market before that date runs to2 December 2026 under the new Article 111(4) of the AI Act, added by the Digital Omnibus on AI (Regulation (EU) 2026/1744 ). That grace doesnot cancel the 50(1) "you are talking to an AI" duty.

If you only remember one engineering takeaway: a chatbot disclosure banner is not the same workstream as watermarking or content marking.

#

When a notice is usually in scope

Ask a product lawyer for your case. From an implementation checklist perspective, teams usually treat a notice as needed when all of these are true:

1. A natural person can interact with the system (chat UI, voice agent, in-app assistant).
2. The interaction is with an AI system, not only a human agent.
3. The person would not obviously know it is AI from the context alone (for example, a feature branded only as "Support" with no AI cue).

Exemptions and "obvious from context" are fact-specific. Do not ship a banner and assume you are done for Article 50 overall. Articles 50(2)–50(4) cover other duties (machine-readable marking, emotion recognition / biometric categorisation notices, and certain deepfake / public-interest labelling rules).

#

A minimal, accessible notice (vanilla JS)

Goal: show a persistent, readable notice before the first message, keep it available to assistive tech, and avoid dark patterns (do not hide the only disclosure behind a one-time toast that vanishes forever with no other cue).

React sketch:

#

Recording which notice shipped

A disclosure you can't evidence later is hard to defend in a review. A tiny, privacy-light log of which notice version rendered on which surface goes a long way. Don't put message content or user identifiers in it.

Bump the version string whenever the copy or placement changes, and keep a short changelog next to it in your repo.

#

Implementation details that usually matter in review

Timing: show the notice at or before first interaction, not only after the model replies. #

Persistence: a single auto-dismiss toast is a weak pattern if nothing else remains on screen that the person was talking to AI. #

i18n: match the UI language the person is using. #

Accessibility:role="note" plusaria-live="polite" helps screen reader users hear the disclosure without stealing focus. #

Logging: keep a product record of which notice version shipped, when, and on which surfaces. That helps later evidence packs; it is not a substitute for legal review. #

Scope: this snippet only targets a 50(1)-style chatbot disclosure. It does not implement 50(2) marking, 50(3) notices, or 50(4) labelling.

#

Limits

• This is not a compliance certification and not legal advice.
• Hosting the notice does not by itself mean Article 50 does not apply in other ways.
• Extra-EU providers can still be in scope when an AI system's output is used in the EU (see Art. 2(1)(c) of Regulation 2024/1689). Confirm with counsel for your distribution model.

#

Next step for teams under time pressure

1. Map every customer-facing AI surface (web chat, mobile, voice, embedded agents).
2. Decide where a 50(1)-style disclosure belongs on each surface.
3. Separately track 50(2) marking work if you generate images, audio, or video for EU users — especially if you are counting on the 2 December 2026 transition for older generative systems.
4. Run a structured scope pass before you file tickets. If it helps, RuleReddy has a free, no-signup Article 50 check that walks through the questions above.

Sources

source & further reading

dev.to — original article Probabilistic Graph Neural Inference for wildfire evacuation logistics networks for low-power autonomous deployments Creator Catalyst: AI-Powered Content Repurposing Platform Eradicating AWS OOM Kills: 10M LLM Inferences on a $0 Cloud-Cost O(1) Memory Clamped Architecture