EU AI Act Transparency Rules for Chatbots and AI-Generated Content Take Effect
How left and right are reading this
- Both agree
- The disclosure duties are now enforceable with real money behind them, bind deployers as well as developers, and are left incomplete by carve-outs and postponed high-risk penalties.
- They split on
- Whether the story is about people finally gaining an enforceable right to know a machine is talking to them, or about firms facing fines under rules clarified one month earlier.
The Facts
- On August 2, 2026, the transparency obligations set out in Article 50 of the EU AI Act began to apply, requiring users to be informed when they are interacting with an AI system and when content has been generated or manipulated by AI.
- AI-generated or manipulated images, audio and video — including so-called deepfakes — must be identified through labels, notices, icons or watermarks, and the requirement also extends to certain texts published to inform the public without prior editorial review.
- Chatbots and other interactive AI systems must clearly disclose to users that they are not human.
- From the same date, the European Commission's AI Office gained new powers to supervise generative and general-purpose AI models and to impose fines for non-compliance.
- Penalties for breaching the new obligations are cited at up to €15 million or 3% of a company's turnover.
- The obligations bind not only developers of AI systems but also the businesses and public bodies that deploy them in services, customer support and automated processes.
- The AI Act was adopted in 2024 and entered into force in August 2024, with its provisions applying in phases; the European Commission published guidelines in July 2026 to clarify how the transparency requirements should be met.
- The new rules include exceptions and grace periods that delay their applicability in certain cases, including for AI systems already on the market, while penalties tied to high-risk AI obligations have been postponed.
- Industry specialists say a central compliance difficulty is that many organisations do not have a full inventory of which AI tools they are actually using.
Context
What exactly has to be labeled now?
Providers and deployers must mark content that has been generated or altered by AI. This covers images, audio and video that resemble real people, objects, places or events, as well as certain texts published to inform the public without prior editorial review Deutsche Welle ZN.UA. Deepfakes must be flagged using icons, notices or specific identifiers Estadão, and identification can take the form of labels, warnings or watermarks Cadena SER. Sources note the rules do not extend to tools performing purely assistive or auxiliary functions ZN.UA.
Who enforces the rules, and what are the penalties?
The European Commission's AI Office begins supervising compliance with several provisions of the regulation from August 2, including for general-purpose and generative AI models, and can impose fines Jornal Expresso infobae. Reported penalty ceilings are up to €15 million or 3% of turnover elEconomista.es ZN.UA. According to one account, member states have until September to adapt their national legislation to the new obligations LaSexta.
Does this apply only to large AI developers like OpenAI and Google?
No. While companies such as OpenAI (ChatGPT), Google (Gemini) and Anthropic (Claude) must design their systems so users know when they are dealing with a machine infobae Diario de Sevilla, the obligations also reach organisations that merely use AI tools in public services, customer support or automated processes ElNacional.cat Expansión. A separate provision, Article 4, has since February 2025 required companies deploying AI systems — including general tools like ChatGPT or Copilot — to ensure staff have a sufficient level of AI literacy ElNacional.cat. Coverage aimed at businesses notes that smaller firms are among those now in scope TechRadar.
Facts first. Then every angle.
The day’s biggest stories in one short brief — the facts everyone agrees on, then the competing values behind the headlines. Free in your inbox.
View all 85 sources
Independent coverage (50)
About these frames
See this differently than someone you know would? Two ways to keep it going.
The dial works on any URL — paste an article you read elsewhere this week.