9

When should AI reveal itself? What Article 50 of the AI Act is really about

On 2 August 2026, Article 50 of the EU’s AI Act came into force, setting out transparency obligations for AI providers and deployers alike. And, as it turns out, its reach extends well beyond high-risk systems.

The new rules go considerably further, covering day-to-day customer service, corporate communications, advertising materials, and even in-house chatbots and voice assistants.

For many businesses, this means having to review their existing processes – and not just on the technology side, but also their editorial and contractual procedures.

Not every use of AI requires disclosure, but the line is not always clear

The obligations under Article 50 depend on three variables:

  • What the AI system does
  • How people interact with the AI
  • What role the organisation plays in the deployment chain

What at first glance looks like a straightforward transparency rule turns out, in practice, to be a set of nuances that may well catch you off guard.

An article written with the help of AI may be published without any labelling, provided it has undergone genuine editorial review and someone actually bears responsibility for its publication. A mere spell-check will not suffice, as the European Commission’s guidelines make very clear. The review must go to substance, and a specific person or organisation must stand behind the published piece.

Conversely, an AI-generated video featuring a real employee, even one who consented to it and approved the script, may still require explicit labelling. This is because approval is not the same as appearing in the video, and the depicted person’s consent does not exempt the video from the labelling requirement.

When a customer interacts with AI – and what follows from that

Systems designed for direct interaction with people must inform users of their artificial nature clearly, unambiguously, and no later than at the point of first contact. A reference buried in the website’s terms and conditions will not suffice. Acceptable options include a brief notice in the chat window or an audible introduction in the case of a voice assistant.

It’s also worth bearing in mind that giving a chatbot a human name and a realistic profile picture makes it considerably harder to argue that its artificial nature was obvious to users.

The exception to the disclosure obligation applies where the interaction with AI would be obvious to a reasonably well-informed and observant person. This assessment must, however, be made from the perspective of how the service is actually presented rather than its underlying architecture.

Article 50 of the AI Act – who is responsible, and for what

Responsibility for meeting the transparency obligations does not always rest with whoever built the system.

An organisation that commissions an AI assistant and deploys it under its own brand may be treated as a technology provider.

Likewise, where AI-generated content is produced on a company’s behalf and under its supervision by an external agency, the commissioning party may bear responsibility for proper disclosure.

The following matters should be addressed in contracts with providers and agencies:

  • Who applies the labels
  • Who preserves metadata during editing and export
  • Who verifies the final version of the publication

A general declaration of ‘AI Act compliance’ will not resolve any of these issues.

The AI Act: what to check now

The transitional period runs until 2 December 2026 and covers only providers’ obligations to label systems placed on the market before 2 August 2026.

The remaining disclosure obligations under Article 50 are already in force, so it’s worth carrying out a practical review of chatbots, advertising materials and published commentary to determine:

  • What role the company plays in each case
  • Whether disclosure or technical labelling is required
  • Who is actually responsible for this

A detailed analysis of individual scenarios, a discussion of the exceptions, and practical implementation guidance can be found in the full version of the article.

Source: BPCC Contact Online

Date: 13 October 2026

Have questions? Get in touch with us


Contact us:

Robert Brodzik

Robert Brodzik

Advocate / Counsel / NewTech / Data Protection and Cybersecurity

+48 532 206 479

r.brodzik@kochanski.pl

Natalia Kotłowska-Wochna

Natalia Kotłowska-Wochna

Attorney-at-Law / Partner/ New Tech, IP, Trade & Logistics Practice Group / Head of New Tech M&A

+48 606 689 185

n.kotlowska@kochanski.pl