Operational guidance, not legal advice. This page distills named public sources (regulator guidance and industry practice). It is not a legal determination, not a notification decision, and not a substitute for your counsel, insurer, or a retained DFIR firm. Verify applicability and current deadlines for your facts and jurisdiction.
How does the public trust buyer chat work?
Last verifiedThe public trust buyer chat at /api/trust/chat is not a generative bot: deterministic keyword routing in src/lib/compliance/trust-buyer-faq.ts runs first; a miss may reach an Anthropic classifier (max_tokens: 12) that returns a topic token mapped to a controlled template — grounded only on the published PublicTrustCenterView.
Trust buyer chat, last verified 11 September 2026 against src/app/api/trust/chat/route.ts, trust-buyer-faq.ts, and trust-buyer-chat.server.ts. Not legal advice.
Audience
Audience: a buyer reviewing a published trust center page, or a vendor security team explaining the widget to procurement. This describes the public endpoint only — not the signed-in app.
Extractive/classifier — not generative
src/app/api/trust/chat/route.ts documents the layered defense: same-origin, rate-limited, resolves a PUBLISHED trust center slug to PublicTrustCenterView, tries answerBuyerQuestion from src/lib/compliance/trust-buyer-faq.ts first, and only on a genuine miss may call buildAnthropicBuyerClassifier from src/lib/compliance/trust-buyer-chat.server.ts.
The classifier in trust-buyer-chat.server.ts uses createOrgAiClient and resolveFeatureModel, calls Anthropic with max_tokens: 12 (one topic token — the model must not write prose), records usage under trust_buyer_chat via src/lib/ai/usage.server.ts, and maps the token to a controlled template. No question is persisted.
This is extractive/classifier behavior — not a generative bot. Answers echo only fields already sanitized into the published trust-center view.
| Property | Behavior | Module |
|---|---|---|
| First path | Deterministic keyword FAQ | src/lib/compliance/trust-buyer-faq.ts |
| Fallback | Topic classifier (max_tokens: 12) | src/lib/compliance/trust-buyer-chat.server.ts |
| Grounding boundary | Published PublicTrustCenterView only | route.ts + trust-buyer-faq.ts |
| Generative prose | No — controlled templates only | trust-buyer-chat.server.ts |
| Question stored | No | src/app/api/trust/chat/route.ts |
| Certification claims | No — echoes published statusLabels + caveat | trust-buyer-faq.ts |
What you need to do now
Publish only what you intend buyers to see — the chat cannot leak unpublished tenant fields because it reads PublicTrustCenterView only.
If your org disables third-party AI, the route returns the neutral miss template without calling the classifier.
Checklist
Verify these items before citing the public trust buyer chat in a review.
- Confirm trust center and compliance modules are enabled for the public endpoint to respond.
- Review what your published PublicTrustCenterView exposes (frameworks, subprocessors, AI governance summary).
- Tell reviewers this is not a generative bot — answers are FAQ routing or template-bound classifier output.
- Note anti-enumeration: invalid slugs get the same neutral miss as unknown topics.