AI Act: Transparency Obligations and Oversight Rules for GPAI Models Now Apply
Key takeaways
Since August 2, 2026, most provisions of the European Union’s Artificial Intelligence Regulation have been applicable. This new phase concerns the transparency obligations applicable to certain AI systems and AI-generated content, the right to lodge a complaint in the event of a breach of the Regulation, as well as the Commission’s supervisory and enforcement powers over providers of general-purpose AI models.
The Artificial Intelligence Regulation (“AI Act”) has been progressively applicable since 2025. (1)
Following a first set of provisions that became applicable on February 2, 2025 (including the obligation to take measures to ensure a sufficient level of AI literacy and the prohibition of AI systems presenting an unacceptable risk), a second milestone was reached on August 2, 2025, with the application of the provisions concerning general-purpose AI (“GPAI”) models, certain governance rules and a significant part of the penalties regime. (2)
August 2, 2026 marks the third major milestone in the AI Act’s implementation timeline. Among the main measures that became applicable on that date are the transparency obligations set out in Article 50, the complaint mechanisms, and the Commission’s supervisory and enforcement powers over providers of general-purpose AI models. However, the effective date of the provisions concerning high-risk AI systems has been postponed.
1. Implementation of the transparency obligations applicable to certain AI systems
The provisions concerning the transparency obligations set out in Article 50 of the AI Act became applicable on August 2, 2026.
As a general rule, the AI systems covered by these transparency requirements should not be used without the individuals concerned being made aware of their interaction with or exposure to AI. The transparency obligation means that providers and deployers of AI systems must inform the users and other natural persons that they are being exposed to or interacting with an AI system.
a) AI systems intended to interact directly with natural persons: providers of AI systems intended to interact directly with natural persons, such as chatbots or voice assistants, must ensure that those persons are informed that they are interacting with an AI system, unless this is obvious from the circumstances and the context of use, and subject to certain exceptions. (Art. 50 §1)
b) AI systems generating synthetic content: providers of AI systems, including general-purpose AI systems, that generate synthetic audio content (such as speech synthesis or artificial voice generation), images, video or text must ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated by AI. (Art. 50 §2)
In addition, specific obligations apply to deployers of certain AI systems.
c) Emotion recognition or biometric categorisation systems: deployers of emotion recognition or biometric categorisation systems must inform the natural persons exposed to those systems of the operation of the system. In addition, the data collected must be processed in accordance with the GDPR. (Art. 50 §3)
d) Deepfakes: deployers of AI systems that generate or manipulate image, audio or video content constituting a deepfake, or text published for the purpose of informing the public on matters of public interest, must disclose that the content has been artificially generated or manipulated. (Art. 50 §4)
This information must be provided in a clear and distinguishable manner, at the latest at the time of the first interaction or exposure. It must also comply with applicable accessibility requirements. (Art. 50 §5)
Certain AI systems, including AI systems intended to detect, prevent, investigate or prosecute criminal offenses, are excluded from these transparency and information obligations.
For businesses, these provisions require reviewing the mechanisms used to inform users, the functionalities used to mark AI-generated content, and the procedures governing the publication and dissemination of such content.
2. Breach of the AI Act: the right to lodge a complaint with a market surveillance authority
A second set of provisions that has become applicable concerns remedies in the event of a breach of the AI Act: any natural or legal person may now submit complaints to the relevant market surveillance authority. (Art. 85)
As a reminder, France has not set up a new market surveillance authority, but has instead designated several existing authorities that will share this responsibility according to their respective areas of competence. For example, compliance with transparency obligations is supervised by the DGCCRF and Arcom. Emotion recognition and biometric categorisation systems are supervised by the CNIL.
3. Deployment of oversight mechanisms for GPAI models
While the obligations applicable to providers of general-purpose AI models have applied since August 2, 2025 (AI Act, Chapter V), the powers enabling the Commission to effectively supervise and enforce those obligations took effect on August 2, 2026.
The Commission and, where appropriate, the AI Office have several supervisory and enforcement tools available in relation to providers of general-purpose AI models:
· the power to request documentation and information. The requested information may be supplied by the provider or by lawyers duly authorised to act on behalf of their clients; (Art. 91)
· the power to conduct evaluations of GPAI models in order to assess compliance of the provider with the Regulation or to investigate systemic risks at Union level associated with the model; (Art. 92) and
· the power to request providers to take appropriate measures to comply with their obligations under Articles 53 (Obligations for providers of general-purpose AI models) and 54 (Authorised representatives of providers of general-purpose AI models), to implement mitigation measures, or to restrict the availability of the model on the market, to withdraw or to recall the model. (Art. 93)
The Commission may impose fines on providers of GPAI models under the conditions set out in Article 101, namely where the provider, intentionally or negligently, has infringed the relevant provisions of the Regulation, failed to comply with a request for documents or information or supplied incorrect, incomplete or misleading information, failed to comply with a measure requested under Article 93, or failed to provide the Commission with access to the GPAI model, including a GPAI model with systemic risk.
The Commission may impose fines not exceeding 3% of the provider’s annual total worldwide turnover in the preceding financial year or EUR 15 million, whichever is higher.
The general application of the AI Act as of August 2, 2026 is subject to one important exception: the Digital Omnibus Regulation on AI amended several provisions of the AI Act and postponed the application of the main provisions concerning high-risk AI systems. (3)
This postponement does not, however, affect the general application of the AI Act as of August 2, 2026. Businesses must therefore distinguish between AI Act obligations that are now fully applicable and those whose application has been deferred. This requires continuing to map their AI systems in order to identify the applicable legal framework and compliance timeline for each of them.
* * * * * * * * * * *
(1) Regulation (EU) 2024/1689 of 13 June, 2024 laying down harmonised rules on artificial intelligence (“Artificial Intelligence Act”).
(2) For the provisions that became applicable on February 2, 2025, see AI Act, Chapters I and II, Arts. 4 and 5; Art. 113(3)(a). For the provisions that became applicable on August 2, 2025, see AI Act, Chapters V, VII and XII, Art. 113(3)(b).
(3) Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July, 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards simplifying the implementation of harmonized rules on artificial intelligence (“Digital Omnibus on AI”).
Bénédicte DELEPORTE
Avocat
Deleporte Wentz Avocat
August 2026