How the roll-up gate works

How the Customer-Facing AI Output Risk Triage gate works

This tool grades every customer-facing AI surface - support bot, product-page copy, FAQ generator - on liability exposure. It reads 4 surfaces and rolls them into one verdict — the worst, never the average — then a gate that can only make it worse.

The 4 surfaces

01SUPPORT-BOTfix firstcontrol 58/100 · policy-invention and human-exit both at 0HIGH RISK

02CHAT-WIDGETcontrol 50/100 · every control only partialREVIEW

03PRODUCT-PAGE-COPYcontrol 85/100 · performance-claim and grounding partialLOW

04FAQ-GENERATORcontrol 100/100 · every control well-controlledLOW

ALL LOW RISK

every surface LOW

SOME TO REVIEW

worst surface is REVIEW - no surface hits the policy-invention + human-exit gate

HOLD THE SET

any surface is HIGH RISK - gated, or its control score falls under 50

Policy-invention + human-exit gate · worsen-only

A surface that can state policy with no grounding (policy-invention control at 0) AND offers no human escalation (human-exit ramp at 0) is HIGH RISK regardless of every other control. The two must fail together - either alone is graded on score alone.

HOLD THE SET

Why: Four customer-facing surfaces, mean control 73/100. Two read clean (FAQ-GENERATOR 100, PRODUCT-PAGE-COPY 85) and CHAT-WIDGET is a middling 50 (REVIEW). But SUPPORT-BOT can state policy with zero grounding and offers no human exit - both gate controls sit at 0. One gated surface is enough: the set posture is HOLD THE SET, worst-not-average. 1 of 4 surfaces HIGH RISK · worst-not-average · mean control 73.

Fix firstSUPPORT-BOT - open the policy-invention or human-exit control (both are at 0); either one releases the gate.

No single surface is the crisis — the clustering is. The roll-up takes the worst, the gate escalates when trouble compounds, and it names the one thing to fix first. Not a live test of your systems and not legal advice. A risk-triage aid; bring counsel to anything it flags.

RedHub AIthe AI that tells you when to doubt it redhub.ai/systems/customer-facing-ai-output-risk-triage