AI disclosure

On this website

This site contains no chatbot, no AI assistant and no system that interacts with you. Nothing you do here is processed by a model. There is no automated decision-making of any kind.

• The logo. Concept exploration for the Verhelm identity used generative image models. The mark that ships is hand-authored SVG, not model output, which is why it scales cleanly and inherits its colour from the page.
• Photography.Every photograph on this site — the brass instruments, the notebooks, the cups of tea — was generated by an image model. None was taken with a camera, and nothing in them is a real place, a real object we own or a real moment. They are set dressing: they show objects, never a client, a person, a place we have worked or a piece of work we have done. Each was produced from a written prompt, and the prompt is published next to the image file — add .prompt.txt to any image address on this site and you can read exactly what was asked for. An earlier version of this page promised to label each generated image where it appears; saying it once here, and publishing the prompts, is a plainer account than a caption repeated under every photograph.
• What is never generated. Anything that shows our work. The specimen assessment on the home page is drawn in code from real clauses and is labelled a specimen twice. We publish no generated screenshots, no invented case studies, no client logos and no photographs of people.
• Written content. Articles and page copy are drafted and edited by the directors. Where an AI tool assists with drafting, a director reviews and is accountable for every factual and legal claim published. Regulatory dates and clause references are checked against primary sources, and each article carries the date it was checked.
• Code. This website was built with substantial help from AI coding tools, reviewed by a human engineer. We mention it because we make the same disclosure a condition of our own build work.

AIAuditSense

AIAuditSense is an AI system and we describe it as one. It uses a large language model, accessed through OpenRouter, to assess evidence you supply against codified framework logic. You are interacting with an AI system, and the following limits are real rather than boilerplate:

• Its output is an orientation tool, not legal advice, and not a conformity assessment within the meaning of the AI Act.
• Classification under the Act depends on facts that are not observable from a website or a repository: how a system is marketed, whose name is on it, what a deployer does with the output, and the contractual chain behind it. AIAuditSense cannot see any of that.
• Language models produce confident text whether or not they are correct. Treat findings as prompts for investigation, not as conclusions.
• Content you submit is retained only as long as needed to produce and revisit your report, with prompt bodies redacted on a 90-day schedule. It is not used to train models.

In client delivery

We use AI tools in our own work, for research, drafting, code generation and analysis. Three commitments govern that:

1. A human is accountable for every deliverable. No assessment, classification, policy or line of code reaches a client without a director having read it.
2. We disclose it on request, and volunteer it where it matters. If a deliverable was substantially produced with AI assistance, we say so before you rely on it.
3. Client data does not go where you have not agreed. Confidential material is processed only through tooling covered by the engagement agreement, and never used to train third-party models.

Our own AI literacy

Article 4 of the AI Act requires providers and deployers to ensure a sufficient level of AI literacy among their staff, and has done since 2 February 2025. It applies to us as much as to our clients. Both directors maintain current knowledge of the frameworks we advise on: the AI Act text and its implementing timeline, the NIST AI Risk Management Framework, ISO/IEC 42001, and India’s DPDP Act and the 2025 Rules under it.

Questions

If something here is unclear, or you think we have understated our use of AI somewhere, tell us. That is the kind of challenge this page exists to invite. Write to us via the contact page.