EU AI Act · Article 50 · transparency obligations

Your AI talks to customers. Article 50 requires it to say so.

The Article 50 Readiness Pack gives you everything you need to comply with the EU AI Act's transparency obligations: the disclosures, the vendor clauses, the evidence, and two executive guides built from the Commission's own material. Self-serve, inside the platform. No consultant.

Applied since 2 August 2026. A permanent obligation, not a deadline that passed.

50days
08hours
26min
12sec

To 2 December 2026, when the Article 111(4) marking transitional ends for generative systems placed on the market before 2 August 2026.

Why transparency is where enforcement starts

Four duties, and the split most organisations get wrong.

Provider
50(1)Say it is AI
50(2)Machine-readable marking
Deployer
50(3)Emotion and biometrics
50(4)Deepfakes and public-interest text
A mark a person cannot see does not inform them. Article 50(5).

Article 50 is the obligation an authority can test from outside your building. It needs no access to your model, your training data or your pipeline: it needs a browser.

The split between provider and deployer is where it goes wrong. Two duties, two duty-holders, and a deployer cannot lean on the provider’s machine-readable marking to discharge its own labelling duty.

§ 50(1) · chatbots and assistants

An undisclosed chatbot is a breach a regulator can screenshot

If a system interacts directly with people and never says it's AI, that's a visible, provable failure. No forensics required, and the Commission's Article 50 guidelines have narrowed the “it's obvious” defence to almost nothing for consumer-facing systems.

§ 50(2) · generated content

Unmarked AI output has to be machine-detectable

Providers of generative systems must mark audio, image, video and text as artificially generated, and make a detection route available. The duty sits on the provider, not across the chain.

§ 50(3) · emotion and biometrics

Emotion and biometric systems must inform the people exposed

A common blind spot in HR, retail and security. The duty to notify sits with the deployer, before exposure.

§ 50(4) · deepfakes and public-interest text

Deepfakes and AI-written public-interest text must be labelled

Deployers must label them clearly and distinguishably, with no transition period. And your supplier's machine-readable watermark does not discharge it. The label has to be one a human can see or hear.

Enforcement

What a market surveillance authority can do without asking.

These are the standard product-safety powers every Member State must confer, and on 2 August 2026 they reached AI systems.

They can buy your product pretending to be a customer

The power to acquire product samples, including under a cover identity, to inspect them and reverse-engineer them, in order to obtain evidence. Regulation (EU) 2019/1020, Article 14(4)(j).

They can turn up unannounced

The power to carry out unannounced on-site inspections, and to enter any premises, land or vehicle used for the business.

They can pull your content

Where the risk is serious enough, order content off your site or a warning displayed to your users, and if you do not comply, require your hosting provider to restrict access to it.

What is in the pack

Six parts, each with an owner.

Accountable: CEO / COO

01 · Scope and role determination

Which of your systems trigger Article 50, and whether you are the provider or the deployer for each. The highest-value step, and the one most organisations have never done.

Accountable: CPO / CMO

02 · Ready-to-paste disclosures

The compliant wording for the §50(1) chatbot notice, the §50(3) emotion and biometric notice and the §50(4) deepfake label, with placement, timing and persistence guidance, and the Commission's own list of approaches that do not count.

Accountable: Legal / Procurement

03 · Vendor and contract clauses

The contract language that puts your AI supplier on the hook to mark output machine-readably, with the acceptance test that proves they did. Where §50(2) is your duty it stays your duty, but the work sits in someone else's pipeline, and this is how you get it done.

Accountable: General Counsel

04 · Code of Practice decision

A clear read on whether to sign the Code of Practice on Transparency of AI-Generated Content: what it gets you, and what it commits you to.

Accountable: CEO / Compliance

05 · Audit-ready evidence pack

The records that show you assessed each system, chose an approach and documented why. Under enforcement the question is “show us what you did.” This answers it.

Accountable: COO / Compliance

06 · Gap report

A scored readiness assessment across the four duties and the Article 50(5) standard, plus the Code decision, with a gap and remediation log, every gap carrying an action, an owner and a target date.

The Article 50 Readiness Pack

Everything the transparency duties ask you to produce.

Self-serve, inside the platform, with no consultant. Buying it starts with a free 30-minute call. It includes everything in the Article 4 AI Literacy Register, so the literacy record and the transparency record are one file rather than two.

See the Article 4 AI Literacy Register

Article 50 Readiness Pack

€1,995One time, plus VAT
Enquire to buy

Includes:

  • Everything in the Article 4 AI Literacy Register, included
  • Scope and provider/deployer role determination
  • Paste-ready §50(1), §50(3) and §50(4) disclosures
  • Vendor and contract clauses for §50(2) marking
  • Code of Practice sign / don't-sign decision
  • Audit-ready evidence pack, plus a scored gap and remediation log
  • Two executive guides: the Article 50 Crosswalk and Article 50, in plain terms
Timing

The dates, and what each one actually reaches.

2 August 2026§50(1), §50(3) and §50(4) applied with no transition at all. §50(2) applied too, for anything placed on the market from that date.
2 December 2026The §50(2) transitional expires for systems placed on the market before 2 August 2026, and two new Article 5 prohibitions begin to apply.
2 February 2027Code signatories commit to an interoperable detection solution.

The transitional is narrower than the commentary suggests.

Article 111(4) reaches providers only, generative systems only, and only systems placed on the market before 2 August 2026. A deployer gains nothing from it.

Partly interactive, partly generative systems get it for one limb only.

The Commission's Guidelines are explicit: such a system benefits from the transitional for the §50(2) marking duty, while the §50(1) interaction disclosure was due on 2 August 2026 regardless.

Retroactivity is asymmetric.

Content generated before 2 August 2026 does not have to be labelled retroactively. The Commission encourages it but does not require it. For images, audio and video the generation date governs. For text published to inform the public on matters of public interest, the publication date governs. Drafted in July and published in August, it needs a label.

Article 50, in plain terms

Every duty, with its exceptions

Who built this

Practitioners, not a template shop.

Written from the Official Journal

Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744, not from somebody else's summary of it.

Checked against the Commission's own material

The Article 50 Guidelines and the Commission's Article 50 FAQ, with the exceptions each duty actually carries.

A formal submission to the Commission

On this Regulation, plus a documented defect in the Commission's own AI Act Compliance Checker.

Find out where you stand, in thirty minutes.

We go through what you run, which duties reach it, and what is outstanding. No obligation, and you keep the findings either way.

Book assessment