AI and Automated Systems Disclosure
1. What this document covers
Trooth uses automated systems to observe, organize and protect the Trooth Network. This document says what those systems do, where generative AI is and is not used, and which decisions are always made by a person. It supplements the Terms of Service, the Privacy Policy and the Acceptable Use Policy. Where this document and those documents differ, those documents govern.
2. Automated systems in the product
Witnessing. When a company connects a system to Trooth, automated collectors read signals from that system on a recurring schedule and record what they observed, with the source and a timestamp. Witnessing observes; it does not modify the connected system. The Connector and Read-Only Access Disclosure describes this in detail.
What a company's page shows. Trooth does not compute or publish a rating, a grade, a rank or any composite figure. What the Network holds for a company is the record of what was witnessed, each observation carrying its source and the time it was observed. An observation is not an AI-generated draft and is never labeled as one; the methodology page describes what an observation records and what deliberately never enters it, and no observation is hand-adjusted to favor or disfavor a company.
Ordering and search. Directory results are ordered by explicit, published criteria: companies with witnessed evidence appear before companies without it, then by the time each record was last witnessed, then by name. Search matches what you typed. There is no paid placement in ranking.
Bot protection. Public forms are protected by Cloudflare Turnstile, which distinguishes people from scripts. A form submission that fails this check is rejected before anything else happens.
Moderation assistance. As the Acceptable Use Policy describes, automated classifiers, anomaly detection and rate-limit triggers can flag content and behavior for review. Flagging is automated; enforcement judgment on borderline cases is not.
3. Where generative AI is used, and where it is not
Trooth-operated AI. Generative AI appears in the product where the AI Use Policy describes it: Kyrie, the assistant that answers questions about your posture, and questionnaire drafting, which proposes answers from evidence Trooth has already witnessed. Every output of these Trooth-operated features is labeled as AI-generated at the point you see it, a person decides what to submit or publish, and the model card describes the models behind these features. Trooth's team also uses generative AI internally for productivity. Trooth-operated AI never publishes a company's declaration.
A company-authorized agent. A company may also use its own browser AI agent, which Trooth does not operate, in the profile editor during an owner's or admin's signed-in session. The agent can change the company's draft, which only the workspace sees. While the workspace's publishing permission is on, which it is by default, the agent may also publish the company's declared fields as the published declaration buyers read, without a person approving each edit: the delegation itself authorizes publication. An owner or admin turns the permission off in Settings, Workspace, and the agent then saves drafts for a person to publish. Each publish by the agent is attributed to the agent in the workspace's audit log, with the person whose session it used. The agent cannot change witnessed evidence, the page address, artwork, the legal-review confirmation or signed model and trust cards. A publish from an editor's session, or with the permission off, is refused as not permitted (HTTP 403), and a change that reads as instructions to an AI model or carries hidden characters is held for review (HTTP 422) until a person publishes it. Nothing is published by a refused change.
Trooth-operated AI does not write the facts on a company's public profile. What a company's own agent publishes is the company's declaration, labeled declared, as if a person at the company had entered it. Witnessed evidence is recorded from observation, not generated; a witnessed fact and an AI-generated draft are never presented as the same kind of thing. Trooth does not use customer content to train AI models, as the AI Use Policy states.
4. Decisions people make
A dispute about a company's page is reviewed by a person, and the disputed party cannot close a case against itself. Corrections and right-of-reply follow the Publication, Correction and Right-of-Reply Policy. Account enforcement follows the review pipeline in the Acceptable Use Policy. Automated systems flag, defer and rate-limit; they do not settle disputes.
5. Accuracy and limits
Automated observation can be incomplete or stale: a connected system can be temporarily unreachable, and a signal reflects the moment it was read. That is why every witnessed fact on the Network carries its source and timestamp, and why the Disclaimer says plainly that Trooth is a witness, not a certifier. If something on a page about your company is wrong, the correction process exists to fix it.
6. Automated access to Trooth
Trooth's pages are served with noai and noimageai robots directives, and Trooth does not authorize the use of Network content to train AI models. Automated access to public data is offered through the documented API instead; the API and Agent Usage Terms govern it.
7. Contact
Questions about this disclosure: legal@trooth.co. Something wrong on a page about your company: use the dispute link on that page or write to support@trooth.co.
Revision note. Revised September 3, 2026. The rating and ordering entries described a composite figure that Trooth no longer computes or publishes. They now describe what a Trust Profile actually contains and how the directory is actually ordered. No disclosure was narrowed, and the correction is recorded under our publication and correction policy.
Revision note. Revised September 20, 2026. One paragraph used "standing" as a name for what a company holds on the Network. It is not a thing a company holds; it is a record of what was observed, and the word invited the reading the rest of this disclosure refuses. The paragraph now says what the page shows. No disclosure was narrowed, and the correction is recorded under our publication and correction policy.
Revision note. Revised September 23, 2026. Spelling is now American throughout, as Trooth is a Florida limited liability company, and dates are written month first. No right or obligation was narrowed, and the correction is recorded under our publication and correction policy.
Revision note. Revised October 4, 2026. Wording only: the word this document used for a single number summing a company up, in its clauses and in the earlier revision notes above, is now "composite figure" or "rating", the words used everywhere else Trooth publishes. The meaning is unchanged: no right or obligation was added, narrowed or removed, and the correction is recorded under our publication and correction policy.
Revision note. Revised October 4, 2026. Section 3 described generative AI as if Trooth operated all of it, said a person decides what to publish, and said generative AI does not write the facts on a company's public profile. Since October 4, 2026 a company's own browser agent, authorized by the company, may publish the company's declared fields without approval of each edit while the workspace's publishing permission is on. Section 3 now separates Trooth-operated AI from a company-authorized agent and states the draft and the published declaration, the owner or admin permission, its default and how to turn it off, attribution in the audit log, the fields the agent cannot change, and how a refused change is handled. The correction is recorded under our publication and correction policy.