EU AI Act Transparency Rules Take Effect, Requiring Deepfake and AI-Content Labeling Across the Bloc
Article 50 of the EU AI Act became applicable on August 2, 2026, requiring chatbot disclosure, machine-readable marking of AI-generated content, and deepfake labeling, backed by fines up to €15 million.
Overview
Transparency obligations under Article 50 of the European Union’s AI Act became applicable on August 2, 2026, requiring companies that build or deploy AI systems in the bloc to disclose when people are interacting with a chatbot, mark AI-generated content so it can be detected as artificial, and label deepfakes and AI-generated text published on matters of public interest, according to the European Commission. The rules of the AI Act “will come into effect in August 2026,” and from that date the EU AI Office and national market surveillance authorities are responsible for implementing, supervising and enforcing the regulation, the Commission’s own overview of the AI Act states.
What the Rules Require
Article 50 sets out four separate transparency duties, according to the Commission’s guidelines on the provision. Providers of interactive AI systems such as chatbots must design them so individuals are explicitly informed whenever they interact with an AI system directly. Providers of generative AI systems must add machine-readable marks to their outputs — text, images, audio and video — so the content can be detected as AI-generated or manipulated. Deployers of emotion-recognition or biometric-categorization tools must inform the people exposed to them. And deployers must disclose deepfakes, along with AI-generated text published on matters of public interest that has not undergone human review or editorial control, per the Commission’s guidelines.
Providers of generative AI systems already on the market before the deadline have an extended window until December 2, 2026, to implement machine-readable marking, according to a legal explainer from artificialintelligenceact.eu. The same explainer notes several exemptions: content that is obvious to a reasonably informed observer, assistive-only functions such as grammar correction, systems authorized for criminal investigation, and artistic, satirical or fictional works, which face reduced rather than full disclosure duties.
Guidance and a Voluntary Code
The European Commission published draft guidelines on the four obligations on May 8, 2026, clarifying how providers and deployers should implement them, according to a client alert from Greenberg Traurig. Alongside the guidelines, the Commission developed a voluntary Code of Practice on Transparency of AI-generated Content, finalized on June 10, 2026 after two earlier drafts, which the Commission and the European AI Board have endorsed as “an adequate voluntary tool” for demonstrating compliance, the Commission says.
By late July, roughly 190 organizations across sectors including IT, telecom, education and retail had signed the code, with 83 signing the providers’ chapter and 152 signing the deployers’ chapter, according to the Commission. Named signatories to the providers’ chapter include Aleph Alpha, Anthropic, Black Forest Labs, Cohere, Google, Meta, Microsoft, Mistral, OpenAI and Synthesia, while the deployers’ chapter counts companies such as Bulgari, Fastweb, Getty Images, Iberdrola, Lenovo and Lufthansa among its signatories, the Commission’s announcement states.
Penalties and Enforcement
Non-compliance with Article 50 can draw fines of up to €15 million or 3% of a company’s total worldwide annual turnover for the preceding financial year, whichever is higher, according to independent analyses from Bratby Law and Greenberg Traurig. EU institutions themselves face a reduced cap of up to €750,000 for breaches, Bratby Law notes. Enforcement responsibility is shared among Member State market surveillance authorities, the EU AI Office and the European Data Protection Supervisor, per the same analysis.
The August 2 date for transparency obligations was not affected by the broader Digital Omnibus negotiations that have reshaped other parts of the AI Act this year. As previously reported by The Machine Herald, the European Parliament and Council reached a provisional deal in May pushing back compliance deadlines for high-risk AI systems to December 2027 and August 2028 — but the general application date for transparency rules and other obligations not covered by those extensions remained fixed at August 2, 2026. Separately, The Machine Herald reported in April that only 8 of 27 member states had designated the national authorities required to enforce the AI Act, raising questions about how consistently the new obligations will be policed on the ground even as the legal deadline itself held firm.
Doubts About Effectiveness
Not everyone is convinced the rules will work as intended. Alexandru Voica of AI video company Synthesia, a signatory to the providers’ chapter of the Code of Practice, told Euronews that “the real issue is a worrying lack of media literacy across large parts of Europe, which makes these technical obligations largely ineffective.” Euronews also reported that no industry-wide detection standard exists, that watermarks can be removed or degraded through editing and compression, and that content created outside the EU can still reach European users regardless of the bloc’s labeling requirements.
What We Don’t Know
How strictly national authorities will enforce the new obligations remains unclear given the enforcement-readiness gaps documented earlier this year. It is also not yet established how regulators will handle content that originates outside the EU but reaches European users, or how consistently the exemptions for artistic, satirical and assistive uses will be interpreted in practice.