Free E-book: AI for the Public Sector Read →
Log in
Free ConsultationFree consultationFree consultationFree consultation

The AI Act in the Czech Republic: New Transparency Rules Effective August 2, 2026

Hello, we’re bringing you the latest legislative updates applicable in the Czech Republic regarding the use of AI.

New transparency rules take effect on August 2, 2026

  • A chatbot or voicebot on a website must be clearly labeled as AI. It must not create the misleading impression that the user is communicating with a human. At the same time, there must be a way to reach a live employee.
  • Texts published by the agency to inform the public about matters of public interest that were generated or significantly edited by AI must be labeled as such. This applies to press releases, website articles, newsletters, and social media posts.
  • The same applies to images, audio, and video that may appear authentic. The CTU recommends labels such as “AI generated” (entirely created by AI) and “AI modified” (edited by AI).
  • If you use tools for emotion recognition or biometric categorization of individuals, you must inform each affected person of this fact.

Practical note: For systems placed on the market before August 2, 2026, the deadline for machine-readable labeling has been extended to December 2, 2026. The rules for high-risk systems will phase out in 2026 and 2027.

Guide from the Office of the Government and NÚKIB: Criteria for Assessing Each Deployment

A lesser-known but often overlooked requirement: According to Section 4 of the AI Act, AI literacy for employees has been mandatory since February 2025 for anyone who uses AI systems in a professional capacity. This includes government agencies and universities.

  • Privacy. Employees must know where the data they enter will go and who has access to it.
  • Transparency. It must be clear what was created by a person and where AI assisted.
  • Responsible use. Prohibit manipulative use, deepfake content, and reliance on unverified outputs.
  • Fairness and non-discrimination. Outputs may be biased even if you explicitly prohibit the AI from discriminating.
  • Security. Use only tools approved by the organization or a verified IT administrator.
  • Preservation of human decision-making. AI must not decide on benefits, permits, or sanctions, nor replace administrative discretion. Formally clicking to approve an output is not sufficient.
  • Legal Compliance. This includes copyright law, administrative regulations, and records management.

A golden rule from the Guide that’s easy to remember: Think of AI as a very eager but inexperienced intern. The signature at the bottom of the document is yours, not AI’s.

The Four Most Common Problems I See at Government Offices

Two things that tend to come as an unpleasant surprise to management:

  1. The government agency is required to keep records of where AI is used, for what purpose, and with what inputs. This does not involve individual employees’ logs, but rather an institutional overview.
  2. If damage occurs due to an unverified AI output, the government agency is liable for it. Compensation can then be sought from the specific employee, and in some cases, even from the AI tool provider.

Six steps you can complete by the end of the year

  1. Identify what is actually being used in your organization. Without this, there’s no point in doing anything else.
  2. Adopt an internal AI policy. Include a list of approved tools, data classifications (public, internal, confidential, sensitive), and a clear list of what should never be fed into AI.
  3. Check your website’s chatbot to ensure it’s labeled as AI.
  4. Implement labeling of AI content in newsletters, on the website, and on social media.
  5. Train your staff. AI literacy is a new skill and, at the same time, a legal obligation. Define roles: who decides on deployment, who owns the process, and who is the IT guarantor.
  6. Address supplier contracts. Specify where data is processed, whether it is used for training, and what the supplier’s legal role is (processor or controller). This year, NÚKIB published guidelines on public procurement in ICT.

For larger-scale processing of personal data, I also recommend conducting a Data Protection Impact Assessment (DPIA).

Sources

  • Guide and Ten Commandments: vlada.gov.cz, Civil Service section, Other Documents
  • CTU Statement on Transparency Rules: ctu.gov.cz
  • Warnings and Lists of Not-Recommended Applications: nukib.gov.cz
  • The Guide also includes useful appendices: an overview of obligations when implementing AI tools from the Ministry of the Interior and a methodology for analyzing suitable use cases from the Ministry of Industry and Trade, including templates.
Free Consultation