WE USE ARTIFICIAL
INTELLIGENCE.
IT DOES NOT USE US,
OR YOU

We work with Claude (Anthropic) under commercial terms. The choice is deliberate: among available providers, Anthropic publishes its safety research, and its commercial terms exclude client content from model training.

We do not submit client material to consumer-tier services, or to tools whose data-processing terms we have not read. This rule applies to our collaborators as it does to us.

CONFIDENTIALITY ALWAYS

Client content is processed only under terms that prohibit its use for model training. Material covered by a non-disclosure agreement is handled in accordance with that agreement; where an agreement prohibits third-party processing, no AI tool is used.

HUMAN WORK IS OWNABLE

Rights in outputs are assigned to us and transfer to the client under our contract. Because purely machine-generated material may not attract copyright protection, human authorship is not decorative: it is what makes the work ownable.

ENVIRONMENTAL IMPACT

AI carries a material environmental cost: greenhouse gas emissions, water consumption for cooling, and the extraction and manufacture of hardware. We cannot isolate the share attributable to a given engagement, and we will not pretend otherwise. Neither can our provider: Anthropic does not publish lifecycle analysis for its models. The reference on this point is Mistral, which in 2025 published — with ADEME and Carbone 4, and under independent audit review — the first lifecycle assessment of a large language model. It is a standard the industry, our own provider included, has yet to meet. What falls to us is proportion: that study establishes that impact scales with model size. We therefore use the smallest model adequate to the task. We claim neither neutrality nor offsetting.

ERRORS ARE OURS

Every deliverable is reviewed, verified and signed by a named person. Errors are ours. The use of AI does not reduce our professional liability in any respect.