AI Policy
The commitments every Carbon Grey Labs product is held to — disclosure, human control, no training on customer content, and honest limits.
Carbon Grey Labs builds AI-native products, and the Studio builds AI systems for clients. Artificial intelligence is central to what our products do: enhancing photographs, drafting listing copy, assembling reports and supporting pricing analysis.
This policy states the principles every Carbon Grey Labs product is held to. It is binding on us. Each product’s own terms describe how these principles apply in that product, and may add to them but may not weaken them.
1. AI-generated content is disclosed
Where our products generate or enhance imagery, text or other content with AI, that content is identified as AI-generated or AI-enhanced so that buyers, marketplaces and other recipients can tell.
We implement this in two ways:
- Machine-readable provenance. Generated and enhanced media carry embedded provenance metadata marking them as artificially generated or manipulated, in a format intended to satisfy Article 50(2) of the EU AI Act (Regulation (EU) 2024/1689).
- Human-visible labelling. Generated imagery is presented with a visible AI-enhanced label by default. Where a product allows that label to be suppressed, doing so is an explicit and recorded act by the user, who takes on the disclosure duty described below.
We do not design products to pass AI-generated material off as unedited photography or human-written records.
What this requires of you
If you use our products in the EEA, Article 50(4) of the EU AI Act places a disclosure obligation on you, as the deployer, when you publish AI-generated or manipulated images, audio or video. That obligation is broader than most people expect: it covers content resembling real objects and places, not only real people — so an enhanced photograph of an actual vehicle or house is within scope.
Our products are built to make that disclosure the default so you satisfy it without extra work. Our Acceptable Use Policy makes it a contractual requirement.
2. People stay in charge
Our products are designed so that a person reviews and approves AI output before it is published, posted or sent. AI drafts; the user decides.
We do not build systems intended to act publicly on a user’s behalf without their review, and we do not use AI to make decisions about individuals that produce legal or similarly significant effects on them.
If you believe an AI output about you was wrong and it mattered, you can reach a human who can look at it and correct it — write to ai-concerns@carbongreylabs.com.
3. Customer content is not training material
Content our customers upload or create in our products — photographs, documents, listings, messages — is used to deliver the service they asked for, and for nothing else.
We do not use customer content to train our own or any third party’s foundation models. We contract with our model providers on terms that prohibit them from training on data we send through their APIs, and we do not permit our providers to retain that data beyond what is needed to return a result and meet their own abuse-monitoring obligations.
Aggregate, de-identified operational metrics — how long a job took, whether it failed — are used to run and improve the service. Those contain no customer content.
4. Accuracy has limits, and we state them
AI outputs are drafts and estimates, not guarantees.
Pricing reports, valuations and market analyses produced by our products are informational tools. They are not professional appraisals, inspections, surveys, or legal, tax or financial advice, and they are not a guarantee of sale price or of any outcome. Verify material facts before relying on them. Each product states its own limits and the main sources it draws on.
Enhanced imagery changes how a real object is presented. It does not change the object. Nothing our products produce is a substitute for inspecting the thing being sold.
5. AI must not be used to deceive
Our tools improve how honest listings look. They are not for making dishonest ones.
Across every product it is prohibited to use AI output to conceal, remove or misrepresent damage, defects, wear, modifications, or the true condition or identity of an asset — for example, editing out accident damage, rust, water damage or structural problems, or presenting a generated image as a photograph of the actual item when it is not.
Doing that is not merely a breach of our terms. Depending on where you are, it may be unlawful misrepresentation, consumer-protection fraud, or a violation of marketplace rules, and the liability is yours. We will suspend accounts we find doing it.
Our Acceptable Use Policy sets out the full list of prohibited uses.
6. Fairness in regulated contexts
Some of our products touch areas with specific anti-discrimination rules — housing above all.
Where a product generates copy for a residential property listing, it is designed not to produce content expressing a preference, limitation or discrimination based on race, colour, religion, sex, familial status, national origin, disability, or any other characteristic protected by the Fair Housing Act, applicable state and local law, or equivalent law in other jurisdictions. This includes the coded and indirect phrasing that fair-housing rules also reach.
We test for this, and we treat a failure as a defect to be fixed, not an acceptable output. If you see one, report it to ai-concerns@carbongreylabs.com.
You remain responsible for what you publish. Review generated copy before it goes out.
7. How we build
We build on established commercial model providers rather than training foundation models ourselves. Where a product draws on external data — public records, market data, marketplace listings — the product says so and identifies the source category.
Before shipping a feature we consider how it could be misused, and we design against the misuse we can foresee. When we find a misuse we did not foresee, we change the product.
We keep a current list of the providers that process customer data at /subprocessors.
8. Regulatory position
We assess our products against the EU AI Act (Regulation (EU) 2024/1689). On our current assessment, no Carbon Grey Labs product is a prohibited practice under Article 5 or a high-risk system under Article 6 and Annex III. The generative features carry Article 50 transparency obligations, which we meet as described in Section 1.
If a product ever moves into the high-risk category — for example, if valuations were offered for use in credit decisions about an individual — we will complete the conformity assessment that requires before shipping it, not after.
9. Raising a concern
If you believe a Carbon Grey Labs product has produced misleading AI content, discriminatory output, or handled your data contrary to this policy, write to ai-concerns@carbongreylabs.com.
We acknowledge reports within 5 business days and tell you the outcome when we have one. We read every report, and we do not retaliate against anyone who makes one in good faith.
10. Changes
We may update this policy. The “Last updated” date reflects the current version. We will not narrow the commitments in Sections 1 to 3 without posting a prominent notice on this page at least 30 days in advance.