AI Act
AI transparency
This page describes how artificial intelligence is used on this site, which systems are involved, on which content, and what remains a human decision.
It answers the transparency obligations of Regulation (EU) 2024/1689 of 13 June 2024 laying down harmonised rules on artificial intelligence (the "AI Act"), whose Article 50 has applied since 2 August 2026. This site is a deployer of general-purpose AI systems supplied by third parties: it neither develops, trains nor places any on the market.
The AI systems in use
| System | Provider | What it does here |
|---|---|---|
| Claude | Anthropic — anthropic.com/claude | Assistance writing the site's texts and its code |
| Nemotron, via NIM | NVIDIA — build.nvidia.com | Drafts of blog articles (French and English versions) and drafts of gear descriptions |
The workflow, content by content
Blog articles. A draft is generated by Nemotron from a topic and an outline I provide. That draft is then read, corrected, completed and approved before publication — never published as it stands. Every published article carries a transparency notice.
Gear descriptions (inventory). Same principle: a generated draft, then reworked by hand. The factual specifications (brand, material, dimensions, how it actually feels in use) are verified or rewritten by a human, because a model cannot know them. Each sheet carries a transparency notice under its description.
Site texts and legal pages. Written with Claude's assistance, then reviewed and arbitrated by me. The legal pages rest on documentary research; they are not professional legal advice.
The site's code. Largely written with Claude's assistance, under human review. This concerns how the site works, not its editorial content.
Translations (French → English). They are human, with no exception. Site content comes from a reviewed, version-controlled translation file; transactional e-mails exist in two hand-written versions, picked according to the recipient's language. The session contract PDF is rendered in French only: a machine-translated contract would carry no weight, and a misreading of a consent or safety clause cannot be undone.
Photographs and images. They are not AI-generated. They are real photographs of the gear and of the sessions documented on this site. No synthetic image is presented as a photograph.
What is never handed to an AI
- No automated decision about a person. No scoring, no ranking, no profiling, no access decision taken by a model.
- No biometric recognition and no emotion recognition. The site analyses no face and no voice for identification purposes.
- No content misleadingly depicting a real, identifiable person — no deepfake, no cloned voice, no synthetic photograph passed off as authentic.
- No visitor's personal data is sent to a model to generate content. What is sent to the providers is the site's text, not its visitors' data.
On that basis, none of the uses described here falls under the high-risk AI systems of Annex III to the Regulation, nor under the prohibited practices of its Article 5.
Human review and editorial responsibility
Every text coming out of an automatic generation goes through a human review before publication. Editorial responsibility remains entirely human: any errors, approximations or omissions that remain are mine, not a model's.
That review is not a formality. Models invent technical details that are plausible and wrong, and on subjects where safety matters — chastity, bondage, restraint — an error would not be harmless. Safety guidance and first-hand accounts are written or verified by hand.
How content is flagged
- A permanent marking sits at the bottom left of every page: it states that this site uses AI, names the models, and links to this page.
- Blog articles and gear descriptions carry a transparency notice next to the text itself, as Article 50 expects — a mention buried in a legal page would not reach the reader where they read.
- Purely functional content (interface, forms) carries no individual notice; the site-wide permanent marking still applies. Gear descriptions do carry one, since they start from a generated draft.
Data sent to the providers
Text sent out for generation passes through the providers listed above and is therefore processed on their infrastructure. Their use of that data is governed by their own terms and privacy policies, which it is up to you to consult through the links in the table above. No machine translation service is called by the site any more.
Visitors' personal data — subscriptions, messages, forms — is processed as described on the Personal data page, and is not used to feed any model.
Questions, corrections, reports
A factual error, content that looks wrongly flagged, a question about how this works: [email protected]. Every reported correction is examined and, where founded, applied.
Last updated: 22 August 2026.