Artificial intelligence lawyer

The AI Act applies in stages: prohibitions on unacceptable-risk practices have been effective since February 2025; transparency obligations for general-purpose AI apply from August 2025; full obligations for high-risk systems enter into force in August 2026; and AI systems already on the market benefit from an additional deadline until August 2027.

The regulation classifies AI systems into four levels: unacceptable risk (prohibited: social scoring, subliminal manipulation, mass facial recognition), high risk (strict compliance, technical documentation and human oversight obligations), limited risk (transparency obligations, notably for chatbots and deepfakes) and minimal risk (no specific obligations, but codes of conduct encouraged).

Providers of high-risk AI systems must: implement a risk management system, ensure the quality and representativeness of training data, maintain detailed technical documentation, ensure transparency and provide information to deployers, enable effective human oversight, and pass a conformity assessment before market placement.

Companies using high-risk AI systems ('deployers') also have obligations: verify the system is used according to its instructions, ensure required human oversight, monitor operation and report incidents, and conduct a fundamental rights impact assessment in certain cases.

Penalties for AI Act non-compliance are significant: up to €35 million or 7% of global turnover for prohibited practices, up to €15 million or 3% for other violations, and up to €7.5 million or 1.5% for inaccurate information provided to authorities.

Under current French law, only a natural person can be recognized as the author of a work. Purely autonomous AI creations are not protectable by copyright. However, when human intervention is sufficiently creative, prompt selection, composition, editing, the person who directed the creative process may claim rights.

We help you structure your creation process to maximize output protection.

Generating deepfakes, hyperrealistic videos or images impersonating someone's appearance, constitutes a violation of image rights and may fall under digital identity theft (Article 226-4-1 of the French Penal Code). The AI Act now imposes a transparency obligation for all AI-generated or manipulated content.

We intervene urgently to obtain content removal and initiate proceedings.

Unauthorized synthetic reproduction of a performing artist's voice violates their neighboring rights (Articles L.212-1 et seq. of the French Intellectual Property Code). This question is at the heart of debates in music law and artist rights. INFLUXIO supports artists in defending their vocal rights against generative AI.

Using copyright-protected works to train AI models raises the question of the text and data mining exception. European Directive 2019/790 provides an exception for research, but its application to commercial uses is contested. Rights holder opt-out is possible but complex to implement. We advise creators and companies on their respective rights.

Algorithms as such are not patentable in Europe, but the software implementing them can be protected by copyright (Article L.112-2 13° of the French Intellectual Property Code). Computer-implemented inventions may also be patented under certain conditions. We define the protection strategy best suited to your situation.

Databases benefit from dual protection under European law: copyright on their structure (if original) and the sui generis right of database producers (Directive 96/9/EC). Your training datasets are a strategic asset whose protection must be anticipated from the collection phase.

The Law of July 30, 2018, transposing Directive 2016/943, provides specific protection for trade secrets. For your AI models, hyperparameters and technical know-how, implementing reasonable protection measures (NDA, access restrictions, traceability) is essential to benefit from this regime.

Pre-trained model usage licenses, API access conditions, usage limitations, reverse-engineering prohibition clauses: each AI license agreement must be tailored to your specific use case and anticipate regulatory developments.

Ownership of rights over developments, deliverables, training data and intermediate models must be clearly defined. We secure your AI development contracts with providers, freelancers and internal teams.

Liability for AI-caused damage is a rapidly evolving topic. The EU AI Liability Directive (Proposal 2022/0303, withdrawn by the European Commission in 2025 due to failure to reach agreement between co-legislators) aimed to facilitate victim compensation. We already integrate limitation and liability-sharing clauses adapted to this emerging framework.