Why regulate AI agents legally?
AIS act autonomously, manage sensitive data, and can have legal implications, but responsibility will be attributed to a human or legal entity.
Therefore, it is necessary to establish a legal framework, including AIS governance, appropriate contractual management with suppliers that meet the requirements of the AI ACT, and an internal organization to ensure the traceability and governance of AIS actions.
A Governance
The company must structure the governance of AI systems to ensure the compliance and effectiveness of AI projects. This can be achieved through the following measures:
I. Master plan
The purpose of this plan is to:
• Set specific objectives for the use of AI (virtual customer service, chat box).
• Specify permitted and prohibited uses.
• Define the acceptable level of risk (e.g., human validation before the dissemination of generated content).
• Integrate the requirements of the GDPR and the AI Act.
• Demonstrate due diligence in the event of audits or litigations.
II. Mapping AI agents and AI risks
This involves identifying the AI agents used and identifying, assessing, and documenting the risks associated with AI by category (legal, reputational, social, or cyber).
III. Internal charter on the use of AI
The charter clearly defines the rules for using AI within the company.
It will mention authorized uses (such as the use of AI to write internal emails with human validation) or prohibited uses (employee performance evaluation without human validation and without transparency on the criteria).
The Charter may require:
- reporting any doubts about AI; respecting transparency and fairness, particularly in external interactions.
- tracking decisions made by high-risk AI systems.
IV. Operational procedures
To ensure compliance with the AI Act, it is useful to establish procedures for:
- Mandatory human validation, particularly for high-risk AI systems (recruitment, medical diagnosis).
- Traceability of decisions: recording AI tools to explain the choices made.
- AI incident management: reporting and analysing any malfunction or challenge to an AI decision.
B Contract management
A contract governing the use of an AI system must be precise and include essential clauses.
I. Legal classification of the AI agent
The contract will specify the legal nature and functions of the AI agent (autonomous assistant, etc.) to avoid any confusion about responsibilities or the risk of inappropriate classification.
II. Distribution of roles and responsibilities
It is essential to identify:
• The supplier (developer or seller of the AI).
• The integrator (responsible for adapting and deploying the AI).
• The operator (responsible for the day-to-day use of the AI).
• The configuration manager (responsible for configuring the AI).
The contract must, in particular:
- specify who is responsible for errors, biases, or damages and define the limitations of liability in order to prevent disputes in the event of an incident.
- stipulate the obligation to comply with regulations (AI Act, GDPR, etc.).
- systematically include regulatory compliance guarantees, including the right to audit, regular transmission of documentation, notification of breaches, and the obligation to cooperate during inspections.
III. Data governance and ownership of outputs
A clear distinction must be made between different categories of data (personal, non-personal, confidential, strategic).
In addition, the contract must establish ownership of incoming data and results, guarantee the absence of bias, respect confidentiality, and protect the intellectual property rights of third parties.
IV. Security and robustness clauses
The contract will stipulate compliance with relevant security standards (including the GDPR), rapid incident notification, mandatory cooperation in the event of cyberattacks, the implementation of enhanced authentication, and continuous monitoring of systems.
Similarly, it will stipulate an obligation for the SIA to be robust.
V. Evolution, updates, and termination of contract
The contract will anticipate the management of SIA versions with the obligation to inform in the event of major changes.
Reversibility solutions will also be provided for (data restoration, transition process with support from the service provider).

