Summary
- The European Commission’s AI Office and national authorities can now enforce applicable AI Act requirements.
- Article 50 introduces disclosure and marking duties for interactive AI systems and generated or manipulated content.
- Providers and deployers must establish where technical marking, user disclosure, and editorial responsibility sit across the AI supply chain.
The European Union has begun enforcing the Artificial Intelligence Act as new transparency rules take effect for chatbots, deepfakes, and other generated or manipulated content.
From 2 August, the European Commission’s AI Office and national authorities can enforce the provisions that have entered application. The change moves the regulation further from legislative preparation into active supervision, complaints, and potential regulatory action.
Article 50 requires providers of systems designed to interact directly with people to ensure users are informed that they are dealing with AI, unless that fact is already obvious from the circumstances and context.
Providers of systems that generate synthetic text, audio, images, or video must also ensure outputs are marked in a machine-readable format and can be detected as artificially generated or manipulated. The technical measures must be effective, interoperable, robust, and reliable where technically feasible.
Separate duties apply to organisations deploying the technology. Deepfakes must be disclosed as artificially generated or manipulated, while AI-generated text published to inform the public on matters of public interest may also require disclosure.
The regulation includes exceptions, including for some law-enforcement uses and for generated text that has undergone human review or editorial control where a person or legal entity accepts editorial responsibility. Artistic, creative, satirical, and fictional works are subject to a narrower disclosure requirement intended not to obstruct the work itself.
The distinction between provider and deployer duties is operationally important. A company supplying a general-purpose model may be responsible for technical content marking, while an organisation incorporating that model into a public service may control how and when the final user is told that AI is involved.
Those obligations become harder to allocate when several companies sit between the model and the final service. A model provider, application developer, systems integrator, and deploying organisation may each control a different part of the content and user journey.
Contracts and technical documentation will need to establish whether machine-readable markings survive editing, compression, conversion, and publication. A disclosure system that works at the point of generation but disappears before the content reaches the public may not provide the transparency anticipated by the regulation.
The Commission has also opened complaint and whistleblowing channels connected to the AI Act, including a route for downstream providers to raise concerns about general-purpose AI models used in their products. These mechanisms create direct paths for governance weaknesses and supplier disputes to reach regulators.
More than 180 organisations have signed a voluntary code intended to help implement transparency requirements for AI-generated content. Participation does not replace compliance with the Act, but the code may influence the technical practices used to demonstrate that generated material can be identified reliably.
The enforcement structure combines EU-level and national supervision. The AI Office holds central responsibilities in areas including general-purpose AI, while national market-surveillance authorities oversee much of the wider market. Cross-border providers may therefore face a common European regulation applied through several supervisory bodies.
UK organisations are outside the EU’s institutional framework, but providers offering in-scope systems in the European market and organisations deploying them within the bloc may still fall within the Act’s reach.
The first enforcement period will test whether AI inventories and compliance programmes reflect how systems are actually used. Organisations will need evidence showing which systems interact with users, which generate public content, how outputs are marked, and who retains responsibility when several suppliers contribute to the final service.
The transparency rules do not apply identically to every use of AI. They do, however, turn technical provenance, user disclosure, and responsibility across the value chain into matters capable of regulatory examination rather than voluntary product design.
