CloudPrimary
A closed frontier model
Claude, through verdict.one's governed gateway. A model that does not train on your clients' data, used only for matters and labels your firm approved.
For solo and small law firms · in development
Use AI on client work — governed.
Your client's data goes only where you have approved, never to a model that trains on it, and every AI action is recorded with what it touched and under what consent.
School policy · research memosynthetic test matter
AI request
- Task
- Summarize · 1 document
- Model
- claude-opus-5-5
- Policy
- Allowed · firm_internal
- Recorded
- Sources, model, retention, consent
Citations in the output · verdict.io
- found347 U.S. 483
- coverage gap612 U.S. 902outside the volumes checked — not a finding of fabrication
How it works
The same AI, with the supervision your bar expects.
The risk is not the cloud. It is a privileged document pasted into a consumer chatbot with no record. verdict.one puts one governed path between your matters and every model.
- 01
Policy before anything is sent
Every matter and document carries a sensitivity. Before any AI call, verdict.one checks it against the providers your firm has approved — and, for privileged work in the cloud, the client consent on file. A denial is recorded too.
- 02
Every AI action is recorded
Which matter, which document versions, which model, under which retention and which consent — written to the matter's audit trail before the request leaves, and closed when it returns.
- 03
Every citation is checked
AI output goes through verdict.io's verification. Each authority comes back found, flagged for review, or outside the covered volumes — and a coverage gap is never called a fabrication.
- 04
A record you can stand behind
A compliance report for each matter states, in plain sentences, what AI touched, where it ran, and what was verified — ready for a court's AI-disclosure order or a client's question.
Where data goes
You decide where each matter's data may go.
Firm-controlledOptional
An open-weight model on your hardware
For sealed matters and clients who forbid third-party processing: the same governance, with inference on a machine your firm controls.
NeverDenied
Models that train on input
Confidential work is never sent to a provider that trains on it unless a client has consented — and that consent is recorded on the matter.
Verification
Checked, and honest about what could not be checked.
A result is either found, flagged for your review, or outside what the checker covers. “Could not verify” never means “fabricated”, and an absence is never clearance. You decide; the record shows what was checked.
separate educational facilities are inherently unequal
Brown v. Board of Education, 347 U.S. 483 (1954) · quotation matched by verdict.io
Ecosystem
Built on the Verdict family, not beside it.
verdict.ioLive
Citation, quotation and AI-output verification behind every check.
verdict.usComing
Alerts when a provision your matter relies on changes — matched on your side, never by sending your list.
Case searchComing
Full-text and semantic search across U.S. case law, inside the matter.
In private development.
verdict.one is being built with a small group of solo and small firms. Early access opens soon.
Built for lawyer supervision, not autonomy · AI output requires attorney review