Four transparency duties, one date that moved
Article 50, the transparency duties, applies since 2 August 2026. The AI Omnibus gave extra time for one part only.
Tell people they are talking to an AI (chatbots, assistants, voice agents), unless it is obvious.
- Who
- Provider
- Applies
- Since 2 August 2026
Mark AI-generated text, images, audio and video in a machine-readable, detectable way.
- Who
- Provider
- Applies
- Since 2 August 2026; 2 December 2026 for systems already on the market before 2 August 2026
Inform people exposed to emotion recognition or biometric categorisation.
- Who
- Deployer
- Applies
- Since 2 August 2026
Disclose deepfakes. Disclose AI-generated text published to inform the public on matters of public interest, unless a person reviews it and takes editorial responsibility.
- Who
- Deployer
- Applies
- Since 2 August 2026
- A chatbot without an AI notice is already late. A product that has generated emails, job ads, summaries or images since before August has until 2 December 2026 to mark its output.
- Breaches of Article 50 carry fines of up to €15 million or 3% of worldwide turnover (Art. 99(4)); for SMEs and start-ups, whichever is lower.
- Emotion recognition in the workplace or in education is not a transparency question: it is prohibited (Art. 5(1)(f)).
What you get
Feature list
Every AI feature that talks to people or generates content, with the Article 50 paragraphs it triggers and why, and the ones that do not, with the reason.
Who does what
For each duty, whether it is yours as provider or your customer's as deployer: what to build, and what to tell your customers.
Disclosure notices
Where and when each notice appears, with draft wording in your product's languages.
Marking
What each generator must embed (metadata, watermark, or both), what your model provider already marks, and what is left for you.
Deadlines
The date that applies to each feature, given when it was first placed on the market.
Transparency pack
A PDF per product, generated from your AI register: everything above, in a form your product team and your customers can read.
How it works
- 01
Kick-off call (30 minutes)
Your products, and which features generate content or talk to users.
- 02
Your input
Screenshots or a demo, product descriptions, and the model providers you use. Optionally, your team runs our review prompt in your own repository.
- 03
Analysis (3 to 4 working days)
Each feature is classified for Article 50 in your register.
- 04
Readout call (45 minutes)
The pack, the open points, and what to build first.
Time and price
- Time
- 1 week
- Price
- Fixed price, agreed before we start
The check is a first step towards the full AI Act audit; its fee is deducted if you go on within three months.
Request a 30-minute callWhy work with us
Your code stays with you
Your team can run our review prompt in its own repository. It asks for a written description of each AI feature, not code, and you read it before anything leaves your environment.
A register, not a slide deck
Every classification is recorded with its reasoning and legal basis in an AI register, and the pack is generated from it, so it stays current when your product changes.
Your data stays protected
We need product documentation, not personal data, and nothing you send is used to train AI models.
A working analysis of your product against the AI Act, not a legal opinion.
Contact
Book a call or send a message about your AI features.
Whether you have a defined project or just a sense that the EU AI Act should be on your agenda — a first conversation costs nothing and usually clarifies a lot.