Responsible AI / Human-in-the-loop Policy

Version de travail opérationnelle — relecture par le propriétaire / un juriste requise. Ce texte est un projet préparé pour relecture et ne constitue pas encore un avis juridique validé. En cas de divergence, la version anglaise prévaut (English version prevails).

1. Human oversight by design

• Every workflow that touches money, customers, hiring or reputation has an explicit human approval gate.
• Low-confidence AI output is held for review, never silently accepted.
• Clients always see where the approval gates sit — no black boxes.

2. Transparency

• Assistants are labelled as assistants; we do not present AI as a human.
• Clients are told which steps use AI and which are deterministic rules.
• Example outputs on this site are labelled as examples.

3. Refusal boundaries

• No workflows that make automated final decisions about people (hiring, firing, credit and similar) — AI output is decision support only.
• No discriminatory criteria, no illegal scraping, no spam, no bypassing platform rules or security controls.
• We decline scopes that require removing human oversight from sensitive decisions.

4. EU AI Act readiness

We track the EU AI Act's obligations and design engagements so clients can meet them: documented human oversight, transparency about AI use, and risk-appropriate controls — particularly in domains the Act treats as high-risk, such as employment.

Tous les documents juridiques