Legal AI TrustCard

TrustCard is an observability scanner for legal AI vendors. You type a vendor name (e.g. Harvey, Legora, Robin AI) and the tool returns a visual scorecard rating that vendor across 8 dimensions of buyer-verifiable transparency: Model Disclosure, System Prompt Visibility, Source Grounding, Audit Logs, Model Versioning, Evaluation Methodology, Data Handling Transparency, and Model Provenance. Each dimension is rated green / yellow / red, with a one-sentence rationale citing the public evidence (or absence of it). The output is a single shareable card buyers can screenshot, send to procurement, or attach to a vendor RFP. TrustCard scores what buyers can observe without taking the vendor’s word for anything.

8-dimension observability scan — Rates vendors on Model Disclosure, System Prompt Visibility, Source Grounding, Audit Logs, Model Versioning, Evaluation Methodology, Data Handling Transparency, and Model Provenance

Evidence-based rationales — Every rating cites specific public signals (developer endpoints, benchmarks, certifications, license files) — not vendor marketing claims

Live AI research — Powered by Claude with web search; reflects current public information at scan time, not a stale annual report

Conservative scoring — Defaults to yellow/red when public evidence is absent; rewards verifiability over vendor self-assertion

Model provenance layer — Surfaces foundation model origin, license, and weight availability via Hugging Face for each model the vendor uses

Shareable scorecard — Single-page visual output designed to be screenshotted, sent to procurement, or attached to an RFP

Buyer-side framing — Scores what buyers can observe without vendor cooperation, not what vendors choose to disclose

No vendor onboarding required — Works on any legal AI vendor from public signals alone; no sales call, no questionnaire, no NDA

Feedback and collaborators