Article 50 transparency disclosure kit
CandidateRank (version 2.4) — Northwind Talent GmbH, HRB 998877, Berlin
For the product, engineering and communications teams implementing Article 50, and for the compliance owner who has to evidence it.
1. What this kit covers and when it bites
Art. 50; Art. 113 second paragraph
Article 50 sits in Chapter IV of the AI Act. It was not caught by the Chapter III deferral in Regulation (EU) 2026/1744, so these obligations have applied since 2 August 2026 — the date on which the Commission's AI Office and national authorities began enforcing the Act. Breaches fall in the Article 99(4) tier: up to EUR 15 000 000 or 3% of total worldwide annual turnover, whichever is higher, and for SMEs whichever is lower.
- Organisation
- Northwind Talent GmbH, HRB 998877, Berlin
- System
- CandidateRank (version 2.4)
- Triggers identified
- interacts, synthetic
- Kit prepared
- 2026-09-11
- Owner
- Jana Novak — Head of People Operations, acting AI compliance owner
2. Interaction disclosure
Art. 50(1)
Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use.
— Art. 50(1)
The duty falls on the provider and is discharged by design, not by a line in the terms and conditions. The "obvious" exception is narrow: judge it from the position of a reasonably well-informed but non-expert user meeting the system for the first time, not from the position of your product team. If you are arguing about whether it is obvious, disclose.
Web or in-app chat — opening message
You're chatting with CandidateRank, an AI assistant from Northwind Talent. It can help with [scope]. It can make mistakes, so please check anything important. To reach a person, type "agent" at any time.
Web or in-app chat — persistent label
AI assistant · Northwind Talent — a person is available on request
Email or written correspondence — footer
This reply was generated by an AI assistant on behalf of Northwind Talent and was not reviewed by a member of staff before sending. Reply to this message to reach a person.
- Place the disclosure at or before the first interaction — not after the first answer.
- Make it visually distinguishable from the conversation itself; an AI-styled bubble that looks like a message is weak.
- Keep it available afterwards as a persistent label, so a user who returns mid-session is not misled.
- Do not bury it in a cookie banner, a tooltip or a link.
3. Machine-readable marking of synthetic content
Art. 50(2); Art. 111(4)
Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.
— Art. 50(2)
This is a provider duty and it is technical, not editorial: the marking must be machine-readable. A visible "made with AI" caption satisfies Article 50(4) for deep fakes but does not satisfy Article 50(2). The Article itself limits the duty where the system performs an assistive function for standard editing or does not substantially alter the input data supplied by the deployer.
| Output type | Marking approach | Owner | Status |
|---|---|---|---|
| Images | C2PA Content Credentials manifest embedded at generation; invisible watermark as a fallback for re-encoding | Jana Novak | [to be completed] |
| Video | C2PA manifest plus container metadata | Jana Novak | [to be completed] |
| Audio | Inaudible watermark plus file metadata | Jana Novak | [to be completed] |
| Text | Provenance metadata where the channel carries it; document properties for generated files | Jana Novak | [to be completed] |
Article 50(7) directs the AI Office to encourage and facilitate a code of practice on detection and labelling of artificially generated or manipulated content; the Commission may give it general validity by implementing act, and may lay down common rules under Article 98(2) only if the code proves inadequate. Adhering to the code is the practical route to demonstrating compliance.
How and when to present all of the above
Art. 50(5)
The information referred to in paragraphs 1 to 4 shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure.
— Art. 50(5)
- Clear and distinguishable: it must not blend into surrounding content or chrome.
- At the latest at first interaction or exposure — never after the user has already acted on the output.
- The information must conform to the accessibility requirements referred to in Article 50(5), so it must work with a screen reader and meet your contrast baseline.
- Article 50 is a floor, not a ceiling: the GDPR, the DSA and consumer-protection law may each require more.
| Trigger | Where implemented | Wording used | Verified by | Date |
|---|---|---|---|---|
| Art. 50(1) interaction | [to be completed] | [to be completed] | [to be completed] | [to be completed] |
| Art. 50(2) marking | [to be completed] | [to be completed] | [to be completed] | [to be completed] |
Implementation owner
Reviewer
Date
What a lawyer should check
Hand this list to counsel with the document. It is short on purpose — these are the points where a generated record most often diverges from the facts of a real organisation.
- Whether any Article 5 prohibition is engaged before relying on the Article 50(3) notice — a notice cannot cure a prohibited practice.
- Whether the "obvious from the point of view of a reasonably well-informed person" exception in Article 50(1) is genuinely available for your interface, or merely convenient.
- Whether your marking approach under Article 50(2) is defensible as state of the art, and whether the Article 111(4) grace period actually applies to this system.
- Whether the human review and editorial responsibility carve-out in Article 50(4) is satisfied in fact, with a named person, and not just asserted.
- Whether GDPR transparency (Articles 13 and 14), Article 22 automated decision-making, or DSA obligations require more than Article 50 does.
- This document is a structured record of an assessment you made, drafted for your review. It is not legal advice and it does not bind any authority.
- It reflects Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744 (the "Digital Omnibus on AI"), in force 27 July 2026. The Commission's guidelines on high-risk classification under Article 6(5) were due on 2 February 2026 and remain in draft (version of 19 May 2026), so classification positions that depend on them may change.
- No harmonised standard has yet been cited in the Official Journal, so the presumption of conformity in Article 40 is not available to anyone.