AI Act Requires Chatbot and Deepfake Disclosure From August 2

Since August 2, 2026, new rules under the European Union's AI Act have been in force, imposing a specific obligation on companies: users must know when they are talking to a bot rather than a human, and any material generated or manipulated by AI must be clearly labeled. This marks the next phase of AI Act implementation, this time focused on transparency toward consumers.

What's Changing

Providers of chatbots, voicebots, avatars and AI agents must clearly inform users from the very first contact that there is no human on the other end. This applies in situations where it would not be obvious to an averagely informed person - a bot with a human name, a profile picture and a natural-sounding voice must be clearly marked as an automated system.

The second pillar of the rules concerns synthetic content. Material generated or manipulated by AI that could be mistaken for authentic must be marked in a visible or audible way, and providers must additionally embed machine-readable markers in files, such as metadata or watermarks.

Deepfakes and Public Content

A particular category is so-called deepfakes, realistic material depicting people, places or events that a viewer could mistake for genuine. A label disclosing the artificial origin must appear no later than the first exposure to the material.

Texts concerning public affairs, such as politics, elections, health or security, are treated separately. If such a text has undergone genuine editorial review by a person responsible for the content, labeling is not required.

The obligation applies to real, two-way and direct communication - European Commission

Exceptions to the Rule

The law provides for several exceptions. Artistic, satirical and fictional content only needs to be labeled in a way that does not interfere with the perception of the work, meaning it does not have to be a large, eye-catching label covering the whole screen. Obvious situations, where the context leaves no doubt about the artificial origin of the material, are also excluded.

According to Dr. Paula Skrzypecka, quoted by rp.pl, a system should inform the user from the very start of the conversation that it is artificial intelligence, without designing the interaction in a way that suggests contact with a human.

The system should inform the user from the outset that it is artificial intelligence - Dr. Paula Skrzypecka

Penalties and Timeline

Violations of transparency obligations carry fines of up to 15 million euros or 3 percent of a company's global annual turnover, whichever is higher. For practices deemed prohibited, such as manipulative use of AI, the fine rises to 35 million euros or 7 percent of turnover.

Companies that placed their generative systems on the market before August 2, 2026 were given additional time, until December 2, 2026, to implement labeling mechanisms. Rules for high-risk systems covered by Annex III of the regulation were pushed back to December 2, 2027, and for systems embedded in other products to August 2, 2028. A new ban on generating sexual material without the consent of the person depicted takes effect on December 2, 2026.

What It Means for Polish Companies

For companies operating in Poland, this means reviewing all customer service bots, virtual advisors and tools generating graphics or video for labeling compliance. In practice, they will need to add a brief notice at the start of a chatbot conversation and a visible marker on images, videos or audio recordings created or processed by AI.

Oversight of compliance in Poland falls to the Komisja Rozwoju i Bezpieczeństwa Sztucznej Inteligencji, though, according to earlier reports, the national oversight structure is still not fully formed, raising questions about how quickly the new obligations will actually be enforced in their first months in force.