The European Commission has updated its FAQs on Article 50 of the AI Act, to supplement the final guidelines published on 20 July 2026. The clarifications specify which organisations must inform users about the use of artificial intelligence and how artificially generated or manipulated content must be marked or labelled. The obligations apply from 2 August 2026.

Providers must design systems intended to interact directly with natural persons in such a way that the user is informed that they are communicating with an AI system. Such a notice is not required only where the artificial nature of the interaction is obvious; this exception must be interpreted strictly in relation to the specific context.

Providers of generative systems, including general-purpose AI systems, must also ensure that audio, image, video and text outputs are:

• labelled in a machine-readable format;

• recognisable as having been generated or manipulated by AI;

• supported by effective, interoperable, robust and reliable technical solutions, within the limits of technical feasibility and the state of the art.

Deployers – that is, organisations that use AI systems under their own authority for professional purposes – must inform individuals exposed to emotion recognition or biometric categorisation systems.

They must also ensure that the following are identifiable:

• images, audio or video that constitute a deepfake;

• text generated or manipulated by AI and published for the purpose of informing the public on matters of public interest.

For the latter texts, the obligation does not apply where the content has undergone human review or editorial oversight and a natural or legal person has assumed editorial responsibility for it. The exemption cannot therefore be invoked on the basis of a purely formal or automated check.

For works that are manifestly artistic, creative, satirical or fictional and contain deepfakes, the information must be provided in an appropriate manner, without compromising the enjoyment of the work. Specific exemptions also remain in place for uses authorised by law for the purposes of preventing, detecting, investigating or prosecuting criminal offences.

The Commission also clarifies that, when the system is used within an organisation, the deployer is normally the legal person. Employees, contractors and freelancers operating under its instructions, responsibility and control do not independently assume that status.

The extension to 2 December 2026 does not apply to all the obligations under Article 50. It applies exclusively to the labelling and traceability of content provided for in Article 50(2) for systems already placed on the market or put into service before 2 August 2026.

The other obligations — including notices regarding interaction with AI, biometric and emotion recognition systems, and the labelling of deepfakes — apply from 2 August 2026. Content already generated and made available before that date, however, does not need to be labelled retrospectively.

Providers and deployers should:

1. carry out an inventory of the chatbots, virtual assistants, agents, generative systems and biometric tools they use;

2. correctly define the organisation’s role in relation to each system;

3. introduce clear and accessible warnings before or at the time of interaction with AI;

4. check with suppliers regarding the availability of machine-readable markings and detection tools;

5. define procedures for labelling deepfakes and content of public interest;

6. document any human review and the effective assumption of editorial responsibility;

7. update contracts, editorial policies, publication procedures and staff guidelines;

8. keep a record of assessments and the solutions adopted to demonstrate compliance.

The Code of Good Practice on the Transparency of AI-Generated Content is a voluntary framework recognised by the Commission and the AI Board to facilitate compliance with labelling and tagging obligations. Adherence to the Code does not automatically prove compliance with the AI Act, whilst organisations that do not adhere to it must be able to demonstrate the compliance of their procedures through appropriate alternative measures.