Trust gradient

The agentic credit broking protocol describes a “trust gradient” between user agents (the AI surface talking to the customer). High-trust user agents can render regulated content inline. Low-trust user agents have to send the customer to a broker-controlled surface for the same content.

The demo implements both. You can switch between them with a URL parameter.

High-trust mode (default)

Disclosures render inline in the customer chat. The customer sees a card with the verbatim text and an “I understand” / “Grant / Refuse” / “I confirm” button. Clicking dispatches the relevant event.

This works because the demo’s user agent is trusted: the broker has audited it, knows what it presents, knows what it doesn’t. In production this is the typical case for a broker’s own white-labelled surface.

Low-trust mode

Append ?trust=low to the customer URL. Disclosures still flow through the same gates and emit the same events, but the rendering changes:

• The inline disclosure card is replaced with a broker-handoff card: “Open on the broker’s secure page →”
• Clicking opens a separate broker-branded page (/broker/disclosure/[sessionId]/[disclosureId]) in a new tab.
• The customer reads the disclosure on the broker surface and clicks the same button.
• The broker page posts the relevant event back to the case.
• The customer chat polls the session every 2 seconds. When the disclosure is acknowledged, the chat continues.

The broker-hosted page has its own visual identity (a deeper amber + slate, with “Broker secure surface” in the header) so it visibly looks different from the inline surface. The customer should be able to tell that they’ve been sent to a different surface.

Why this matters

The protocol’s §7.3 and §7.4 describe two patterns:

• High-trust user agent + verbatim inline rendering = compliant, on the basis that the broker has audited the user agent and trusts it not to corrupt the regulated content.
• Low-trust user agent + broker-controlled surface = compliant on the basis that the regulated content is rendered by a surface the broker fully controls, even though the surrounding journey runs through an untrusted user agent.

The same journey can support both. New user agents (third-party AI assistants the broker hasn’t audited yet) start as low-trust and get promoted to high-trust after audit.

The two modes side by side

Best demo move: open the same session in two browser tabs, one with ?trust=high and one with ?trust=low. Walk a stakeholder through them in parallel. They’ll see:

• Same eligibility check on both
• Same indicative quote on both
• Same application details form on both
• Different disclosure rendering at the status, credit-search-consent, and pre-contract gates
• Same audit log at the end, with the same events recorded regardless of which surface the customer used

The demo controls bar (?demo=1) has a Trust dropdown for switching mid-journey.

Implementation note

The broker-hosted page is at /broker/disclosure/[sessionId]/[disclosureId]/page.tsx. It’s standalone, doesn’t import the customer chat, has its own colour palette via styled-jsx. In production it would live on a separate domain owned and operated by the broker.

The polling pattern in the customer chat (every 2s when trust === "low" && there's an open disclosure) is the simplest possible synchronisation. In production this would be a server-sent event or websocket subscription instead.

Customer experience

From the customer’s perspective:

• High-trust: “I’m reading a disclosure inline, I tap I understand, the chat continues.”
• Low-trust: “I’m being sent to the broker’s secure page in a new tab, I read the disclosure there, I tap I understand, I close the tab and the chat continues automatically.”

The low-trust mode is slightly more friction for the customer (a tab switch). The trade-off is that the broker has full control of how the regulated content is rendered, which is the requirement when the user agent is unknown.

When to choose which

User agent	Recommendation
Broker’s own white-labelled customer journey	High-trust
Audited third-party user agent on a published allowlist	High-trust
New third-party user agent in pilot	Low-trust until audit
Customer’s general-purpose AI assistant (e.g. someone asking a chatbot for a loan and being routed)	Low-trust
Voice-first surface (Alexa, Siri integrations)	Low-trust, with audio-rendered version of broker-hosted page

The demo doesn’t try to dictate the policy. It demonstrates that both modes work end-to-end with the same underlying state machine.

See Adoption path for brokers for the policy considerations.