THE TRUST LAYER

The UPL two-lane guard — built in, not bolted on.

Information vs. advice, enforced before any output ships. It runs on every surface — Firm Brain, monitoring, contract review, the GC line, and white-label client portals.

ClassifyEnforceLog
Information lanegrounded, citation-backed → delivered
Advice laneescalated to attorney review → gated
1
Classify
Information or advice?
2
Enforce
Allow · Block · Escalate
3
Log
upfirm.lane_events

Runs on every surface before output. A compliance safety net — not a replacement for attorney review.

CLASSIFY

Detect advice signals before they ship.

Every input and output is scored for advice-boundary signals. The classification, its confidence, and the matching signals are recorded — not just a pass/fail.

app.upgpt.ai/dashboard/upfirm
Lane classifier scoring an output for advice-boundary signals with a confidence bar and matched signals

ENFORCE

Allow, block, or escalate to the attorney lane.

The enforcement decision is deterministic and explainable, and always carries the standard information-not-advice disclaimer.

Allow

Pure legal information

Grounded, citation-backed — delivered with the standard disclaimer.

Escalate

Advice-boundary content

Routed to the attorney lane for review before it reaches the client.

Block

Out-of-scope / high-risk

Withheld; the event is logged with the classifier rationale.

AUDIT

Every decision logged by matter.

The lane-events log captures the classifier score, the rationale, and the source excerpt — scoped to the matter, ready for a compliance review.

app.upgpt.ai/dashboard/upfirm
Matter-scoped lane-events audit log showing allowed, escalated, and blocked decisions with classifier scores

ETHICAL WALLS

Matter scoping and cross-matter isolation.

Retrieval and outputs stay inside their matter boundary, paired with your firm’s existing ethical-wall policies.

app.upgpt.ai/dashboard/upfirm
Matter isolation view showing per-matter access lists, ethical-wall screens, and a conflict-of-interest banner

GET STARTED

Show compliance how the guard works.

Bring your risk partner to a 30-minute walkthrough — we’ll trace a real query end-to-end through classify, enforce, and log.