Part of the AI Regulation News hub.
The Czech Telecommunication Office tells firms to settle whether they are a provider or a deployer before the Article 50 transparency rules bite on 2 August 2026
The Czech telecom regulator did not write new law. It wrote a sorting exercise, and the sorting is where most companies will get Article 50 wrong.
Bottom line: A press release, not an instrument. It creates no obligation of its own and binds nobody. The obligations it describes come from Article 50 of the AI Act, which the regulator says takes effect on 2 August 2026.
Who this affects: Compliance officers and general counsel at Czech chatbot and voicebot operators, marketing and communications teams publishing AI-generated audio, image, video or text, newsroom editors publishing AI-assisted text on matters of public interest, and any employer running emotion recognition or biometric categorisation.
Issue date: Dated Prague, 31 July 2026. The date the release points at is 2 August 2026, when the Article 50 rules take effect. No consultation deadline, because it is not a consultation.
What changed: Nothing in the law changed on 31 July. What changed is that the Czech market surveillance side of Article 50 now has a public position, including CTU naming itself as one of the Czech supervisors and flagging that the European AI Office will enforce Article 50 instead in some cases.
Analysis: The release spends its first substantive paragraph on role determination rather than on labels. That is the tell. Article 50 splits duties unevenly between providers and deployers, and CTU notes that a company running its own AI system carries both sets at once.
Primary sources: CTU press release, 31 July 2026 (Czech) · Commission guidelines on Article 50 transparency obligations · Code of Practice on Transparency of AI-generated Content
- Instrument (EN)
- Press release: 2 August 2026 as the key date for the transparency rules under the Artificial Intelligence Act. What does it mean in practice?
- Authority
- Cesky telekomunikacni urad (CTU), the Czech Telecommunication Office
- Jurisdiction
- Czechia, applying EU Regulation 2024/1689 (AI Act)
- Status
- Published guidance in the form of a press release
- Bindingness
- Not binding. The underlying Article 50 duties bind; this reading of them does not.
- Issue date / next deadline
- Issued 31 July 2026. The date it flags is 2 August 2026.
- Supervisor named
- CTU as one of the Czech market surveillance authorities, with the European AI Office competent in some cases
- Language
- Czech only, with links to two English-language Commission documents
- Primary source
- https://ctu.gov.cz/tiskova-zprava-2.-srpna-2026-jako-klicove-datum-pro-pravidla-transparentnosti-podle-aktu-o-umele
The three steps, in the order CTU put them
CTU recommends proceeding in three basic steps. Determine your own role correctly. Then identify the specific obligations that apply to you. Then set up the processes and measures that deliver compliance.
Order matters here. A company that starts with labelling technology and works backwards to the legal question will label things it had no duty to label and miss things it did.
On the regulator's account, Article 50 sets duties for the provider of an AI system, meaning the person who places it on the market or puts it into service, and for the deploying entity, meaning the person who uses an AI system in an activity other than a personal non-professional one. CTU adds the case people forget: where a person uses their own AI system, that person must meet both the provider duties and the deployer duties.
Provider duties: interaction, then machine-readable marking
Article 50(1) requires providers to ensure that natural persons are appropriately informed that they are communicating with a chatbot, a voicebot or another system intended for direct interaction with natural persons.
Article 50(2) is broader than most summaries suggest, and CTU says so explicitly. The provider duty to ensure outputs are properly marked in a machine-readable format, and so detectable, covers all audio, image, video or text content, subject to the stated exceptions.
That is a system-level engineering duty rather than a publishing duty. It sits with whoever ships the model or system, and it does not wait for someone to publish the output.
Deployer duties are narrower, and that is the point
By contrast, the deployer duty under Article 50(4) reaches only deepfakes and text published for the purpose of informing the public on matters of public interest.
So where image, audio or video content is created or significantly altered by a deployer using AI in a way capable of creating an impression of authenticity, a duty arises to inform the public that the content was created or altered with AI. The same holds for text generated or manipulated by an AI system and published to inform the public about matters of public interest.
Article 50(3) is separate again: deployers using AI for emotion recognition or biometric categorisation of natural persons must properly inform every affected natural person of that use.
What a compliant label looks like, per the regulator
Article 50(5) calls for clear and distinguishable ways of marking. CTU gives worked examples in English: labels or icons of the type "AI generated" for fully generated content, or "AI modified" for content edited with AI.
Those are the regulator's own examples, and it points readers to the Code of Practice on Transparency of AI-generated Content for further examples of suitable marking.
CTU also reminds readers that the AI Act carries a set of exceptions, for instance for purposes connected with the detection of criminal activity, and recommends that affected parties assess whether their activity falls under one of them.
Who will actually be knocking
Supervision of these duties falls to the market surveillance authorities of the relevant EU member state. In Czechia, CTU says, that will include CTU itself.
In some cases the European Office for Artificial Intelligence will enforce Article 50 instead of those national authorities. Which body is competent within Czechia depends on who the provider or deployer is, and where the AI Office is competent, also on the provider of the general-purpose AI model the system is built on.
For a Czech company that means the answer to "who regulates us" can differ system by system, and can turn on a supplier relationship it does not control.
What we did not verify
We opened the CTU press release at ctu.gov.cz in full, in Czech, and every fact and the pull quote above come from it. Its own dateline reads Prague, 31 July 2026.
We did not open the two documents CTU links to, the Commission guidelines on Article 50 transparency obligations and the Code of Practice on Transparency of AI-generated Content, and we make no claim about what they say beyond CTU's characterisation. We also did not check the Czech AI Act adaptation legislation to confirm how CTU's supervisory competence will finally be allocated in Czech law.
We will not claim that CTU has been formally designated as a Czech market surveillance authority under the AI Act. The release states that CTU will be among the supervising bodies; that is a statement of the regulator's own expectation, not a citation to a designating instrument, and we did not find one.
Article 50 is not one duty, it is four, and they land on different parties. Before your team argues about watermarking, write down for each AI system whether you are its provider, its deployer, or both. On CTU's reading, a company using its own system is both, which is the combination that produces the longest obligation list and the one most compliance registers get wrong.
Source File
Open the CTU press release and confirm the Prague dateline of 31 July 2026, the three-step sequence beginning with role determination, and the passage stating that the Article 50(4) deployer duty reaches only deepfakes and text published to inform the public on matters of public interest.
CTU doporucuje vsem subjektum, ktere vyvijeji nebo pouzivaji systemy umele inteligence, aby se s novymi povinnostmi seznamily a vyhodnotily, zda se na ne jednotlive pozadavky vztahuji. (CTU recommends that all entities developing or using AI systems familiarise themselves with the new obligations and assess whether the individual requirements apply to them.) ยท Cesky telekomunikacni urad press release, 31 July 2026
FAQ
Does this press release create any new obligation in Czechia?
No. It is a press release from a regulator and binds nobody. The obligations it describes are those of Article 50 of the AI Act. CTU's contribution is a reading of them plus a statement about who supervises.
If we only use a third-party chatbot, do we have to watermark its output?
The machine-readable marking duty in Article 50(2) is described by CTU as a provider duty, sitting with whoever places the system on the market or puts it into service. A deployer's own duty under Article 50(4) is narrower, covering deepfakes and text published to inform the public on matters of public interest. If you also run your own AI system, CTU says you carry both sets.
What wording does CTU suggest for a label?
It offers two English examples: "AI generated" for fully generated content and "AI modified" for content altered with AI. It points to the Code of Practice on Transparency of AI-generated Content for further examples, and cites Article 50(5) for the requirement that marking be clear and distinguishable.
Who enforces Article 50 against a Czech company?
The market surveillance authorities of the relevant member state, which in Czechia will include CTU on its own account. In some cases the European AI Office enforces instead, and CTU says the allocation can depend on the provider of the general-purpose AI model underlying the system.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.