Réglementation de l'IA

Requirements depend on jurisdiction, activity, affected people, and sector. A voluntary framework, a proposed rule, and an enacted law are different kinds of documents.

Points clés à retenir

• Identify the use and jurisdiction.
• Distinguish law, guidance, and proposals.
• Reassess when the system or its purpose changes.

Start with the actual use rather than the label AI. A system used for ordinary drafting raises different questions from one involved in employment, credit, healthcare, or public services. Identify where the organization operates and where affected people are located. Separate binding obligations from guidance and proposals. Record the issuing authority, document status, effective date, and relevant scope. A news article about a proposed requirement does not establish that the requirement is currently in force. Map responsibilities across the system. A model provider, application developer, deploying organization, and data supplier can have different roles. Contractual allocation of work does not automatically remove an organization’s applicable obligations. Use a documented review process for changes in purpose, data, providers, and deployment regions. Keep records of decisions, evaluations, and incidents where appropriate. For a real compliance determination, use current authoritative texts and qualified advice for the specific facts. This guide explains how to organize the inquiry rather than certifying a product as compliant.

1. Imagine a vendor citing a proposed AI rule as proof that its product meets all legal requirements.
2. Identify the proposal’s issuing authority, current status, jurisdiction, and the activity it covers.
3. Separate the vendor’s actual controls from its legal claim, and obtain a fact-specific review before relying on the claim.

The hypothetical example avoids treating a proposal or marketing statement as a completed compliance assessment.