EU AI Act: businesses urged to prepare for AI-output and deepfake labelling duties

Amsterdam-based Jeroen Schouten and Nienke Kingma of Pinsent Masons, who specialise in AI regulation and technology law, made the recommendation ahead of transparency rules under the EU’s AI Act taking effect on 2 August.

Under Article 50 of the AI Act, providers of AI systems face a general duty to ensure people they intend to interact directly with their system are informed that they are interacting with an AI system. That duty does not apply if it is “obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use”. Exceptions also apply for certain law enforcement purposes.

Providers must also ensure that, where their system generates synthetic audio, image, video or text content, those outputs are marked in a machine-readable format and detectable as artificially generated or manipulated.

Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, must also generally disclose that the content has been artificially generated or manipulated, subject to limited law enforcement-related exceptions. If the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the transparency obligations “are limited to disclosure of the existence of such generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work”.

Businesses must also generally tell people when they are using an emotion recognition system or a biometric categorisation system.

The various transparency obligations must be discharged “in a clear and distinguishable manner at the latest at the time of the first interaction or exposure” and the information must “conform to the applicable accessibility requirements”. The information must be made available no later than before the first interaction with or exposure to the content.

The European Commission published a code of practice to support compliance.

Schouten said: “This voluntary code of practice provides practical implementation of the requirements for marking AI-generated content and labelling ‘deep fakes’. It sets out the technical and organisational steps providers and users of gen-AI systems can take to comply with the transparency requirements. Organisations that sign the code of practice may also be better placed to demonstrate to customers, end-users, and other stakeholders how they meet the obligations.”

Kingma said: “Businesses should not treat these transparency obligations as a purely technical labelling exercise. They should map where AI-generated or AI-manipulated content is produced or used across the organisation, assess which Article 50 obligations apply, and update governance, product, marketing and user-facing disclosure processes accordingly..”

EU lawmakers recently finalised amendments to the AI Act that will have the effect of delaying the deadline for complying with the transparency obligations for some businesses. For gen-AI providers whose gen-AI models are released before 2 August, they will have until 2 December this year to meet the new requirements. The requirements – including in respect of deep fake disclosures – otherwise apply from 2 August, Kingma highlighted.