AI Transparency

Eight commitments govern every use of AI at Meridian Atelier. The clauses below say how each one is kept.

• Directed. AI does only the work the Principal assigns, within the limits he sets.

• Accountable. The Principal reviews everything AI makes for a client or for publication, and is responsible for all of it.

• Disclosed. This page says how the firm uses AI, every AI image on this site but the small portrait beside the Principal's name carries the AI mark and says so in its description, and an AI that speaks to you says it is an AI.

• Purposeful. Every image of a design starts from a written brief that says what it must show and why.

• Own work. No brief names another designer, artist or brand, or asks for their style.

• Truthful. AI shows designs, materials and options. It never invents a client, a review or a finished project, and shows no invented faces.

• Private. No image on this site shows a client's home. Nothing a client shares with the firm is published without their consent, or used to train an AI model.

• Current. The firm reviews this standard when its tools, its practice or the relevant law change, and the date at the top of this page says when it last changed.

AI does four kinds of work at Meridian Atelier, each under the Principal's direction.

• Images. MAI makes this site's pictures of homes, rooms, materials and places with AI image models, chiefly Google's Gemini, each from a written brief (clause 05). The Principal's portraits are real photographs of him, enhanced with AI (clause 06).

• Writing. MAI, the firm's AI system, drafts the pages of this site (this one included), the write-up of the Principal's reasons for each design choice, and its reading of each client's Diagnostic, the study of how a household lives that the firm makes before any design begins. The Principal reviews and approves every word before it is published or sent.

• Knowledge. MAI maintains the Meridian Engine, the firm's written body of knowledge on designing and building a home, and keeps it current. It also checks each draft against the firm's written rules for its work.

• This website. MAI writes the code that builds this site and draws its diagrams, such as the compass on the Meridian Engine's page.

MAI runs on AI services from other companies: Anthropic's Claude, with OpenAI's ChatGPT in a supporting role. Model training is turned off on all three services the firm uses (Claude, ChatGPT and Gemini), so the companies behind them do not use what the firm shares with them to train their AI models.

Five parts of every project stay with the Principal.

• The design. The Principal designs every client's home; AI does not.

• The decisions. The Principal makes every decision about your home, together with you.

• The approval. Nothing AI makes or drafts is published or sent until the Principal has approved it.

• The responsibility. The Principal is responsible for the drawings and documents your home is built from.

• The relationship. You work with the Principal himself. MAI drafts some of his messages to you; he reviews, approves and sends each one himself. MAI does not write to clients on its own.

The firm shows its designs in AI images rather than in photographs of built homes, for three reasons.

• Clients' homes stay private. The firm does not publish its clients' homes.

• The options are too many to photograph. The materials, finishes and forms the Meridian Engine sets out combine into more homes than could ever be built, and a photograph needs a built home.

• Options can be compared. You can see one kitchen in two stones, or one room in four color palettes, before anything is ordered or built.

An image shows what a design is meant to look like. It is not a photograph of a built home, and the firm never presents one as a photograph, a client's home, a named project or a real address. The homes in these images are not designs for a client: they illustrate the firm's principles and options. MAI drafts each brief from the firm's knowledge, the Principal approves it, and the image model fills in what the brief does not set, such as furnishings and small features.

No image starts from a general request such as “a beautiful living room.” Each image of a design starts from a written brief that names four things. MAI drafts the brief to the firm's rules for images, and the Principal approves it.

• What it is for. The point the image makes on the page where it appears, such as how a stone takes light or how a plan's rooms connect.

• What it shows. The room, each material and finish, and the palette, each color by its name in the firm's own catalog of colors.

• How it is seen. The camera's height and lens, the time of day, and the single light source and its direction.

• What it leaves out. Text, brand names and invented faces (a close view of craftwork may show a pair of hands), and anything that could not be built as shown.

Images are generated only on the Principal's instruction, and he chooses which one is used. A result that misses its brief is corrected one change at a time, or its brief is rewritten. An image of a real place, such as a neighborhood the firm works in, is an AI impression of its character, not a photograph, and is made to show no identifiable property.

• The mark. Every AI image on this site carries the AI mark: a few small four-point stars in a circle, set just inside the image's lower-right corner (a film's top-right), in Bone White over a dark picture and Oxblood over a light one. Point at it to read what it marks; choose it to open this page. The one exception is the small portrait beside the Principal's name (see Portraits).

• The description. Every AI image on this site has alternative text (the description a screen reader speaks and a search engine reads), which ends with its label: (AI-generated) for an image MAI made with an AI image model, or (AI-enhanced) for a real photograph enhanced with AI.

• The watermark. Images made with Google's image models also carry Google's invisible SynthID watermark, which Google's tools can read to identify them as AI-made.

• Portraits. The Principal's portraits are real photographs of him, enhanced with AI for a professional presentation, and are labelled AI-enhanced: the person in them is the Principal himself. Shown as pictures, they carry the mark. The small portrait beside his name on an article or a page carries none, because his name stands beside it.

• Diagrams. Diagrams such as the compass on the Meridian Engine's page are drawn by code MAI writes, not by an image model. They explain an idea in lines and show no real place, so they carry no mark and their descriptions carry no AI label.

• Writing. Pages do not mark which passages MAI drafted. MAI drafts much of the writing on this site, this page included; the author named on a page has reviewed and approved every word and is responsible for it.

MAI works with your project's records, your Diagnostic among them, through the AI services named in clause 02, to prepare the Principal's drafts. Model training is turned off on each of those services, so your information is not used to train their AI models.

How the firm keeps and shares your information is set out in the Privacy Policy. Questions about it go to the Principal, by email or phone at the foot of this page.

This standard draws on the public rules and frameworks below: two are US law, one is European Union law that can reach firms outside the EU whose AI output is used there, and three are voluntary frameworks.

• The American Institute of Architects. Its guidance on the responsible use of AI by architecture and design firms (source, opens in a new tab) (2025): tell clients when and how AI is used, have a qualified professional review every result, and keep the professional accountable for all of it.

• The European Union. Its AI Act (Regulation (EU) 2024/1689, Article 50 (source, opens in a new tab)): people must be told when they are dealing with an AI, unless it is obvious, and AI images that resemble real people, places, objects or events and could pass as authentic must be disclosed clearly, at the latest when a person first sees them.

• California. Its bot law (Business and Professions Code § 17941 (source, opens in a new tab)): no one may use a bot online to mislead a person in California about its artificial identity in order to push a purchase or sale or to influence a vote; whoever uses a bot is not liable if they disclose, clearly and conspicuously, that it is a bot.

• The US Federal Trade Commission. Its rule on consumer reviews and testimonials (source, opens in a new tab) (2024) bans fake or false reviews and testimonials, including AI-generated fakes and ones credited to people who do not exist; the firm publishes no such review. The Commission holds claims about AI to the same law as any other claim a business makes (Operation AI Comply (source, opens in a new tab), its 2024 action against deceptive AI claims).

• NIST, the US National Institute of Standards and Technology. Its voluntary AI Risk Management Framework (source, opens in a new tab) (2023) sets out what makes AI trustworthy, including being valid and reliable, safe, accountable and transparent, privacy-enhanced, and fair with harmful bias managed.

• The OECD, the Organisation for Economic Co-operation and Development. Its AI Principles (source, opens in a new tab) (2019, updated 2024), a non-binding intergovernmental standard, include transparency and explainability, accountability, robustness, security and safety, and respect for human rights, including privacy and fairness.

This page describes how the firm works today. To ask whether AI was used in something you saw or received from the firm, or to question anything here, write or call: the Principal will answer.

The firm reviews this standard when its tools, its practice or the relevant law change, and the date at the top of this page says when it last changed.