All articles
AI for business8 min read

Transparency obligations under Article 50 of the EU AI Act: What industrial and MedTech companies must comply with from 2 August 2026

Transparency obligations under Article 50 of the EU AI Act: What industrial and MedTech companies must comply with from 2 August 2026

Since 2 August 2026, providers and deployers of certain AI systems have had to comply with the transparency obligations in Article 50 of the EU AI Act (Regulation (EU) 2024/1689). For industrial companies and MedTech manufacturers, this is their first direct point of contact with the AI Act, regardless of whether their actual AI applications already fall under stricter rules.

“From now on, everything created with AI has to be labelled.” You have been reading this sentence for weeks. It is wrong. Article 50 does not contain a blanket labelling obligation. It sets out four distinct cases with different addressees and several exceptions. This article shows what that means in practice for an industrial company and for a MedTech manufacturer, with examples from both sectors.

What does Article 50 of the EU AI Act require?

Article 50 sets out four transparency obligations: disclosing AI interaction, marking AI-generated content, informing people about emotion recognition, and labelling deepfakes and certain AI texts.

The four cases, each with the party responsible:

  1. Disclosure of AI interaction (para. 1: providers): AI systems that interact directly with people must be designed so that the person recognises they are communicating with an AI. Exception: when this is obvious from the context.
  2. Marking of synthetic content (para. 2: providers): Providers of generative AI must mark generated image, audio, video and text content in a machine-readable format as artificially generated or manipulated.
  3. Information about emotion recognition (para. 3: deployers): Anyone using a system for emotion recognition or biometric categorisation must inform the people affected.
  4. Labelling of deepfakes and AI texts of public relevance (para. 4: deployers): Deployers must disclose artificially generated or manipulated image, audio and video content, as well as AI-generated texts on matters of public interest.

Provider or deployer: who has which obligation?

Providers develop or supply AI systems and bear the obligations in paras. 1 and 2; deployers use them under their own authority and bear the obligations in paras. 3 and 4.

Both industrial companies and MedTech manufacturers are deployers in most cases: they rarely develop their own generative models, but they use AI systems bought in from others – chatbots, text and image generators, sometimes emotion recognition. For them, the deployer obligations come first, while the technical marking under para. 2 is the responsibility of the provider of each tool. That is why it is important to coordinate with the providers you use: does their system already mark generated content in a machine-readable way, or do you need to ensure disclosure yourself?

What does Article 50 mean for industrial companies?

For industrial companies, Article 50 mainly concerns AI that faces customers and the public: chatbots in technical support, AI-generated product and marketing texts, and AI images.

Concrete cases from everyday industrial practice:

  • Chatbot in technical customer service: Falls under para. 1. Users must recognise that they are writing to an AI and not to a service technician.
  • AI-generated product descriptions, datasheet texts or blog articles: If such texts are published on topics of public interest, para. 4 applies – unless they go through a genuine editorial review with clearly assigned responsibility.
  • AI images in marketing and sales: As synthetic content, they fall under the marking and disclosure logic of paras. 2 and 4.

A purely internal AI knowledge system, on the other hand – for example for research in technical documents, with no direct contact with people and no published content – does not usually trigger a labelling obligation under Article 50.

What does Article 50 mean for MedTech companies?

For MedTech manufacturers, Article 50 concerns AI that faces customers and patients, not the clinical AI in the medical device itself, which is regulated separately as high-risk AI.

This distinction is crucial, because the two are often mixed up in practice:

  • Patient- or customer-facing chatbots, for example on the website or in the support portal, fall under para. 1: disclosing the AI interaction is mandatory.
  • AI-generated content in communication, training or patient information falls under paras. 2 and 4, depending on the context, with the editorial exception for reviewed texts.
  • The AI function in the medical device itself, such as a diagnostic algorithm, is not a case for Article 50. It is subject to the high-risk regime of the AI Act and its interplay with the MDR or IVDR, for which market surveillance in Germany lies with the BfArM (Federal Institute for Drugs and Medical Devices). That is a separate area of work and not the subject of the transparency obligations.

For MedTech manufacturers, this means: from 2 August 2026, the transparency obligations are the part that must be met immediately and independently of product approval – wherever AI interacts directly with people or produces published content.

Do I really have to label all AI content?

No. Article 50 provides for several exceptions: for art and fiction, for editorially reviewed content, and for purely assistive or non-substantial editing.

The most important forms of relief for both sectors:

  • Editorial review: AI texts on topics of public relevance are exempt if they are subject to human review with clearly assigned responsibility. Simply “skimming through” is not enough.
  • Non-substantial editing: Systems that only assist with inputs or process them without substantially altering them are exempt.
  • Art, satire, fiction: Disclosure must not impair the presentation of the work.
  • No retroactive effect: There is no obligation to label deepfakes created before 2 August 2026 after the fact.

Check each case individually instead of labelling everything as a precaution. Unnecessary notices do not create certainty; they only dilute the places where labelling really matters.

Deadlines, fines and responsibility

The transparency obligations have applied since 2 August 2026 and are enforceable from that date; a transition period until 2 December 2026 only concerns the technical marking of generative systems already on the market.

From 2 August 2026, the phase of active enforcement begins, and the competent authorities take up market surveillance. Infringements of Article 50 can be punished with fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher; for SMEs, whichever is lower. In Germany, the Bundesnetzagentur (Federal Network Agency) is responsible for central supervision and runs an AI service desk; for medical devices with AI, market surveillance remains with the BfArM.

What should industrial and MedTech companies do now?

Record the AI systems in use, identify the Article 50 cases that apply, and set up notices, responsibilities and evidence properly.

Four steps:

  1. Create an AI inventory. Record which AI applications are in use and which of them generate content, interact with people or recognise emotions.
  2. Clarify your role. Are you the provider or the deployer of each system? MedTech manufacturers should draw a clear line between customer- and patient-facing AI (Article 50) and the clinical AI in the product (high-risk regime).
  3. Implement notices and processes. Write clear AI notices for chatbots and published content, define responsibilities, and adapt workflows as well as contracts with agencies or freelancers.
  4. Keep evidence auditable. Document which content is labelled and how, and where the editorial review takes place.

This last point is exactly where MAIA comes in. Whether it is technical documentation in mechanical engineering or regulatory documents in MedTech: answers backed by sources and a complete audit trail make disclosure and labelling verifiable. MAIA delivers answers that can be traced down to the document version and page; customer data is not used for training, and the platform is designed for European data protection and compliance requirements, Made in Germany, Switzerland & Austria. For an industrial or MedTech company that has to document its AI use and keep it auditable, this is the foundation on which the obligations under Article 50 can be met in practice.

FAQ

Since when have the transparency obligations under Article 50 of the EU AI Act applied? Since 2 August 2026. For the machine-readable marking of generative systems that were already on the market before then, a transition period applies until 2 December 2026.

What does Article 50 mean for industrial companies? It mainly affects chatbots in technical support, AI-generated product and marketing texts, and AI images. Purely internal AI knowledge systems with no contact with people and no published content do not usually trigger a labelling obligation.

What does Article 50 mean for MedTech companies? It affects AI that faces customers and patients, such as chatbots and published AI content. The clinical AI in the medical device itself does not fall under Article 50, but under the high-risk regime of the AI Act in conjunction with the MDR or IVDR (market surveillance by the BfArM).

Do I have to label all AI content? No. Article 50 covers four distinct cases. Exceptions apply to art, satire, fiction and editorially reviewed content.

What fines can be imposed for infringements of Article 50? Up to €15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs, whichever is lower.

About MAIA

MAIA is the AI knowledge platform for industry and MedTech in the DACH region. It unlocks technical and regulatory documents, checks requirements against internal documentation and delivers answers that can be traced down to the document version and page – exactly the traceability that Article 50 and the documentation obligations behind it require in practice. Customer data is not used for training, and the platform is designed for European data protection and compliance requirements, Made in Germany, Switzerland & Austria. Would you like to see how this helps you document your AI use and keep it auditable? Book an intro call.

Sources

Bundesnetzagentur. (2026). KI-Servicedesk zur Umsetzung des EU AI Act [AI service desk for implementing the EU AI Act]. Bonn: Bundesnetzagentur.

European Union. (2024). Regulation (EU) 2024/1689 (AI Act), Article 50. Official Journal of the European Union.

Fraunhofer Academy. (2026, July). Transparenzpflicht nach Art. 50 KI-Verordnung [Transparency obligation under Art. 50 of the AI Act]. https://blog.academy.fraunhofer.de

IHK Schleswig-Holstein. (2026, July). Neue Transparenzpflichten ab August 2026 – was Unternehmen jetzt prüfen sollten [New transparency obligations from August 2026 – what companies should check now]. https://www.ihk.de/schleswig-holstein

TÜV Rheinland Consulting. (2026). Transparenzpflichten im EU AI Act – Artikel 50 [Transparency obligations in the EU AI Act – Article 50]. https://consulting.tuv.com

usd AG. (2026, July). KI-Omnibus: Die neuen Fristen des EU AI Acts [AI omnibus: The new deadlines of the EU AI Act]. https://www.usd.de

More articles

Let’s get started.

In a first intro call, we’ll work out together what a good start with MAIA looks like for you.

Cookie settings

We store your choice for 12 months.

Necessary

Security, forms and storing your choice.

Always on

Marketing

Measures whether our Google ads lead to enquiries. Google Tag Manager, Google Ads. No remarketing.

privacy policy