EU AI Act Article 50 Takes Effect: AI Disclosure Failures Can Cost 3% of Global Turnover

The practical effect of these rules can be tested over the next three to six months. When an EU user opens a chatbot, AI agent, or avatar, the service should identify itself as AI at the start of the interaction. When a user downloads synthetic images, audio, or video, visible labels and machine-readable marks should remain detectable. If major services merely rewrite their terms while leaving interfaces and files unchanged, the implementation of Article 50 will have fallen short of the legal text.

The European Commission put the AI-system transparency rules into effect on 2 August 2026. Providers and deployers must tell users when they are interacting with an AI system rather than a human; the Commission specifically lists chatbots, AI agents, and avatars. The timing matters because this is a product-interface obligation, not a disclosure that can be buried only in a privacy policy or technical document.

Labels depend on the content and its use

Certain AI-generated or manipulated material must carry a clear, visible label as well as a machine-readable mark. The Commission’s scope includes images, audio, and video that resemble existing people, objects, places, entities, or events, including deepfakes. AI-generated text published to inform the public about matters of public interest also requires disclosure when it has not undergone human review or editorial control. Text that has been reviewed by a human editor should not automatically be treated as meeting that same condition.

The transparency framework also covers emotion-recognition and biometric-categorisation tools. A company therefore needs to inventory what each system does and how it is used before deciding where a notice must appear and which outputs require marking. The two sources do not quantify the engineering hours or compliance cost for different products, so they do not support a comparison between the burden on a small developer and that on a large platform.

The European Commission has issued implementation guidelines, a set of icons for labelling AI-generated content, and a voluntary code of practice that companies can use to demonstrate compliance. A visible label helps a person recognise synthetic material. A machine-readable mark allows platforms and tools to inspect provenance. These mechanisms serve different audiences, and omitting either one would not fully match the Commission’s description of the obligation.

€15 million turns disclosure into an operating risk

National market-surveillance authorities handle general enforcement. The European AI Office supervises systems within its remit, while the European Data Protection Supervisor covers EU institutions acting as providers or deployers. For companies, a transparency violation can bring a maximum fine of €15 million or 3% of worldwide annual turnover. EU institutions, bodies, and agencies face a ceiling of €750,000, and enforcement must take proportionality into account for small and medium-sized enterprises and small mid-cap companies.

Euronews reports that the AI Office must oversee firms with far greater resources while competing for scarce technical talent and keeping up with rapidly changing models. A rule becoming enforceable therefore does not mean every service can be audited at once. The report also notes that foreign companies are covered when they commercialise AI technologies in the EU. In the short term, this may change release processes: providers need to make model capabilities and limitations understandable to downstream users, and some European launches could arrive later while compliance work is completed.

The AI Act entered into force on 1 August 2024 and applies through a staged timetable. The immediately observable change here is Article 50 disclosure and marking; it should not be conflated with every high-risk-system rule, model-documentation duty, or prohibited practice. The next measurable evidence will be investigations and fines announced by national authorities, the placement of notices in EU versions of major services, and whether a machine-readable mark survives after synthetic content leaves its original platform. Those records will show whether implementation has moved from guidance into products.

Sources:

• European Commission: Safer and more transparent AI
• Euronews: EU rules on AI models become enforceable. What’s going to change?