AI Act Compliance Lawyer: anticipate your obligations and secure your AI systems.
The European Regulation on Artificial Intelligence (AI Act) governs the development, placing on the market and use of artificial intelligence systems in the European Union.
Having entered into force in August 2024, it is rolling out its obligations progressively until it becomes fully applicable in August 2027.
Whether you are a provider, deployer, importer or distributor of an AI system, the AI Act imposes obligations on you, and failure to comply can result in fines of up to €35 million or 7% of total worldwide annual turnover.
Betty Jeulin Avocat helps you understand your obligations and bring your AI systems into compliance.
Your questions about AI Act compliance
When does the AI Act come into force?
The European Artificial Intelligence Regulation (AI Act) is entering into force in phases.
• 2 February 2025: Prohibition of unacceptable-risk AI practices (Article 5) and AI literacy obligations (Article 4). Already in force.
• 2 August 2025: Obligations relating to general-purpose AI (GPAI) models. Designation of national authorities. Entry into application of the sanctions regime. Already in force.
• 2 August 2026: Application of the majority of obligations, high-risk AI systems (Annex III), transparency obligations (Article 50), and registration in the European database. Next major deadline.
• 2 August 2027: Full application, including high-risk systems integrated into regulated products (medical devices, toys, etc.).
• 2 February 2025: Prohibition of unacceptable-risk AI practices (Article 5) and AI literacy obligations (Article 4). Already in force.
• 2 August 2025: Obligations relating to general-purpose AI (GPAI) models. Designation of national authorities. Entry into application of the sanctions regime. Already in force.
• 2 August 2026: Application of the majority of obligations, high-risk AI systems (Annex III), transparency obligations (Article 50), and registration in the European database. Next major deadline.
• 2 August 2027: Full application, including high-risk systems integrated into regulated products (medical devices, toys, etc.).
Who is affected by the AI Act?
The AI Act applies to all operators in the AI value chain: providers (developers), deployers (user companies), importers, and distributors of AI systems. Its application is extraterritorial: any company whose AI system produces effects within the EU is affected, even if it is established outside of Europe. This includes startups, SMEs, large corporations, and public administrations.
What practices are prohibited by the AI Act?
Since 2 February 2025, Article 5 of the AI Act has prohibited eight categories of AI practices deemed to pose an unacceptable risk: subliminal manipulation that alters human behaviour; the exploitation of vulnerabilities related to age, disability, or socio-economic situation; social scoring (algorithmic social rating); predictive policing based solely on profiling; untargeted scraping of facial images from the internet or CCTV; emotion recognition in the workplace and in educational institutions (except for medical or safety reasons); biometric categorisation that infers sensitive data; and real-time biometric recognition in public spaces for law enforcement purposes (subject to strict exceptions).
Must I indicate that content has been generated by AI?
The AI Act distinguishes between two situations:
(i) For deepfakes (audio, video, or image content that appears to realistically depict real people or events), labelling will be mandatory from 2 August 2026.
(ii) For AI-generated text on matters of public interest, information regarding the artificial origin of the content must also be provided, unless there is human editorial control. Outside of these cases, there is currently no general obligation to mention that content is AI-generated, although sector-specific obligations may apply (such as in advertising or journalism). The legislative trend is clearly moving towards greater transparency.
(i) For deepfakes (audio, video, or image content that appears to realistically depict real people or events), labelling will be mandatory from 2 August 2026.
(ii) For AI-generated text on matters of public interest, information regarding the artificial origin of the content must also be provided, unless there is human editorial control. Outside of these cases, there is currently no general obligation to mention that content is AI-generated, although sector-specific obligations may apply (such as in advertising or journalism). The legislative trend is clearly moving towards greater transparency.
Does the labelling of AI content apply to artistic works?
The AI Act includes a specific provision for AI-generated artistic content: labelling can be carried out in a way that does not alter the artistic nature of the work. This nuance allows creators a degree of flexibility in how they provide notice, for example by inserting the information into the file's metadata or in a dedicated section, rather than directly on the work itself.
What are the transparency obligations for AI model training data?
Since August 2025, providers of general-purpose AI (GPAI) models have been required to provide technical documentation that includes a sufficiently detailed summary of the training data used. This obligation is intended to allow copyright holders to verify whether their works were used to train an AI model and, if applicable, to exercise their right to opt out. Providers must also implement a policy to respect copyright.
Are companies that use a general-purpose AI model subject to transparency obligations regarding training data?
The obligations regarding training data documentation primarily fall on providers of GPAI models (including the retraining of existing models). However, companies that integrate a general-purpose AI model into their own AI system (such as a chatbot, analysis engine, or decision-making tool) assume the responsibilities of a deployer under the AI Act. They must ensure the overall compliance of their system, which includes verifying the technical documentation provided by the general-purpose AI model provider.