AI policy

This is a translation of the French original, which governs in case of divergence.

We use artificial intelligence tools in the work we deliver, within a defined frame, and your data does not enter them without your written agreement.

Where AI comes in#

Our software engineering work is assisted by generative AI tools, coding agents built on large language models among them. They speed up the writing of code, tests and documentation. We use the same tools for internal work: research, drafting, analysis. We do not hide the practice; this site and the platform behind it are themselves its output. An image shown on a page of this site that a generative model made or changed carries an "AI" mark, which opens to "AI-generated" or "AI-edited" under the pointer or a touch, and its file states the same in its metadata; the illustrations of our case studies are of that kind.

What does not change#

Responsibility for what we deliver is ours, entirely. Everything an AI tool produces goes through human review before it ships, and the same quality bar applies whatever the tool: automated tests, code review, continuous integration, style and security rules. A deliverable leaves our hands not because a model generated it, but because an engineer checked it and stands behind it.

Your data#

We submit no data, code or document a client entrusts to us under an engagement to an AI tool without that client's prior written agreement. Where that agreement exists, we work only with vendors on professional terms that contractually exclude training their models on what is submitted; we check that clause in each vendor's contract before we adopt it, and we keep a copy of it. Confidentiality obligations and the data processing agreement outrank any convenience of tooling.

The permission to name a client in a case study (name, logo, quote, figures) is separate from the above: it covers publishing those elements and is granted in writing for that specific use. Drafting that case study with the help of our AI tools falls under our general practice described above, since the published text is neither data, nor code, nor a document the client entrusted to us for their engagement.

A deliverable an AI tool produced in whole or in part follows the intellectual property regime of article 10 of the terms and conditions: the assignable rights on that deliverable are assigned to the client. Content an AI tool generated may carry no copyright at all, for want of a human author; that is an absence of a right, not a right we would keep for ourselves.

What we do not do#

We take no fully automated decision producing legal or similarly significant effects on people. We deliver no AI-generated content without human validation. We conceal no use of AI tools on an engagement: if you want to know how they took part in your project, the question is welcome and the answer is documented.

The EU AI Act#

Under regulation (EU) 2024/1689 (the AI Act) we are the deployer of the AI systems we use under our own authority; the general-purpose AI models those systems rest on are supplied by third parties, and the regulation keeps the two notions apart. We supply neither a high-risk AI system nor a general-purpose AI model ourselves. Article 4, as amended by regulation (EU) 2026/1744 (the AI Omnibus), in force since 27 July 2026, requires us to take measures supporting our staff's AI literacy; no level is guaranteed. The list of AI vendors used on an engagement is available on request. Where an engagement consists of designing or integrating an AI feature for a client, each party's roles and obligations under that regulation are set out in the engagement's contract.

Questions#

This policy will move with our practices and with the regulatory frame. Questions: info@avanserv.com.