Our chatbot went live: the European AI Act chatbot checklist under Article 50
A live chatbot that talks to natural persons falls under Art. 50 of Regulation (EU) 2024/1689. Art. 50(1) requires the provider to design it so that the person is informed that they interact with an AI system; Art. 50(5) fixes the timing and form. Where the chatbot also generates synthetic text, Art. 50(2) adds a machine-readable marking duty. Art. 50 applies from the general application date In force, which Regulation (EU) 2026/1744 did not move. This page gives the checklist as a sequence, the evidence it leaves and the consequence of inaction.
Where this starts
The chatbot shipped with a disclosure line drafted for the interface, and nobody has read that line against the article. The first question is role. A company that develops the chatbot, or has it developed, and offers it under its own name or trademark is the provider under Art. 3(3), and the Art. 50(1) design duty is a provider duty. A company that uses a vendor's chatbot under its own authority is the deployer under Art. 3(4); the deployer duties in Art. 50 sit in paragraphs (3) and (4), which reach a chatbot only where it runs emotion recognition or biometric categorisation, or publishes AI-generated text on matters of public interest. A company that builds the chatbot and runs it in its own product holds both roles at once. The role tests run first in the sequence below; the generative AI hub carries the obligation set with its two dates.
The second question is whether the disclosure line meets Art. 50(1) and Art. 50(5). The exemption in Art. 50(1) applies where the interaction is obvious to a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use; the published Commission guidelines on Article 50 transparency obligationsPublished read that as an average-person test the provider assesses and records per function. Art. 50(5) requires the information to be clear and distinguishable, to be given at the latest at the time of the first interaction, and to conform to the applicable accessibility requirements. A notice that lives in the terms of service, a menu or a help page is the placement the guidelines do not accept, and it is one of the findings listed against Art. 50(1) in the Art. 50 obligation record. The notice has to be observed in the interface as a new user sees it; a policy statement does not record where the notice sits.
The third question is whether the chatbot also generates synthetic text within the meaning of Art. 50(2). Art. 50(2) addresses the provider of an AI system that generates synthetic audio, image, video or text content, and requires the output to be marked in a machine-readable format and detectable as artificially generated, with a technical solution that is effective, interoperable, robust and reliable as far as technically feasible. The duty does not apply to the extent that the function performs an assistive function for standard editing or does not substantially alter the input data or its semantics, and the guidelines give examples for that exemption. Art. 50 sets no numeric threshold and no JTC 21 draft covers it; no harmonised standard is cited in the Official Journal for Art. 50, so no presumption of conformity is available for any technique. The level of the source hierarchy in use is a Commission code and guideline: the published Code of Practice on Transparency of AI-Generated ContentPublished names the marking layers and, for free-form text, a watermark-only layer above a declared minimum length. A chatbot that generates synthetic text and was placed on the market before the Art. 50 application date has a grace period for the Art. 50(2) marking duty only Deferred; the interaction notice under Art. 50(1) has no grace period. Which technique set counts as adequate under Art. 50(2) is an unsettled point.
The sequence
The work runs in the order below: role and function assignment first, then the legacy status, then the notice observations, then the marking check where the chatbot generates text, then the record.
Determine whether the organisation is the provider, the deployer or both for the chatbot.
Record Role determination record: provider and deployer tests for the chatbot
[Regulation (EU) 2024/1689, Art. 3(3) and 3(4); PR-CLS steps 8 to 9]
Assign the conversation function and each generating function to the Art. 50 paragraph that applies.
Record Article 50 applicability record, one row per chatbot function
[Regulation (EU) 2024/1689, Art. 50(1) to (4); PR-CLS step 35]
Observe each function with a claimed exemption and compare its behaviour with the exemption condition.
Record Article 50 exemption observation record, one row per claimed exemption
[Regulation (EU) 2024/1689, Art. 50(1) and 50(2); PR-CLS step 36]
Record whether the chatbot was placed on the market before the Art. 50 application date.
Record Art. 111(4) legacy status record
[Regulation (EU) 2024/1689, Art. 111(4); PR-CLS step 37]
Start an interaction on each interface as a new user and record when and where the notice appears.
Record interaction-notice-log and notice-placement-log per interface
[Regulation (EU) 2024/1689, Art. 50(1) and 50(5); PR-50-MARK steps 26 to 27]
Run the prescribed accessibility checks on the notice and classify it as clear, distinguishable, timely and accessible.
Record accessibility-check-log and disclosure-classification per notice
[Regulation (EU) 2024/1689, Art. 50(5); PR-50-MARK steps 33 and 40]
Where the chatbot generates text, record the target, run the text detector and compare the layer set.
Record target-justification-record, text-detection-log and layer-set-comparison
[Regulation (EU) 2024/1689, Art. 50(2); PR-50-MARK steps 6, 16 and 38]
Write the audit record per interface and list each gap with its Art. 50 paragraph.
Record Interaction disclosure and label audit record and remediation list (PR-50-MARK output)
[PR-50-MARK steps 43 to 44]
Evidence this situation ends with
| Artifact | Answers | Produced by |
|---|---|---|
| Article 50 applicability and exemption determination record (PR-CLS output) | Art. 50(1) and 50(2): the paragraph assigned to each chatbot function and the observed behaviour against each claimed exemption | PR-CLS v1.0 |
| Record of the assessment that an interaction is obvious to a reasonably well-informed, observant and circumspect person | Art. 50(1): the exemption assessment for each function where the provider relies on it | the provider |
| Interaction disclosure and label audit record (PR-50-MARK output) | Art. 50(1) with 50(5): the moment the notice appears, its location in the interface, and its classification as clear, distinguishable, timely and accessible | PR-50-MARK v1.0 |
| Marking architecture description per output modality | Art. 50(2): the marking layer applied to generated text, the pipeline point where it is applied and the detector that reads it | the provider |
| Marking presence and survivability test report (PR-50-MARK output) | Art. 50(2): the text detector result per sampled output at the declared minimum length, survival after copy and paste, and the layer set compared with the published Code | PR-50-MARK v1.0 |
| Grace-period record for a system placed on the market before the Art. 50 application date | Art. 111(4): the steps taken to meet Art. 50(2) and the date each step completed, for a chatbot that was already live | the provider |
| Remediation list (PR-50-MARK output) | Each gap with its Art. 50 paragraph, the observed behaviour, the record reference and the re-test criterion | PR-50-MARK v1.0 |
Each artifact follows a specification with a synthetic sample on the deliverables page: the applicability record and the audit record follow the assessment memo specification, the marking report follows the test report specification, and the remediation list follows the findings register specification. The test checklist generator lists the Art. 50 test items for an LLM system type with the protocol that runs each one.
If nothing is done
Art. 50 has applied since the general application date, and a chatbot that went live without the notice is in breach of Art. 50(1) from its first interaction with a natural person, with no grace period. Enforcement sits with the market surveillance authority of the Member State under Art. 74; where the authority evaluates the chatbot under Art. 79(2) and finds that the Art. 50 duties are not met, it requires the provider to take corrective action, to withdraw the chatbot or to recall it within a period the authority sets. A breach of the transparency obligations of Art. 50 falls in the penalty tier of Art. 99(4)(g), and the amount in a given case is set by the national authority under the Art. 99(7) criteria. The action on the provider's side is the design change Art. 50(1) describes: the notice moves into the interface at the first interaction, is observed there as a new user sees it, and the observation is recorded; where the chatbot generates synthetic text and was on the market before the Art. 50 date, the marking steps under Art. 111(4) complete within the grace period and the grace-period record states each step. Where the chatbot is also a high-risk system, Art. 50(6) keeps the Chapter III requirements in force alongside.
Services, guides and tools for this situation
The Article 50 transparency and marking/watermark testing service runs PR-50-MARK v1.0 on the notice and, where the chatbot generates text, on the mark, within the limits stated on what we cannot do; the role and function assignment that precedes it is a PR-CLS v1.0 output of the classification and readiness assessment.
- Obligations
- Guides
- ICP hubs
Every inquiry receives a reply within 24 hours. Scope and availability are discussed per request.