For legal

AI-assisted work that survives a privilege challenge

Legal teams do not need permission to use AI; they need a record. Shield lets counsel use frontier models on contracts, discovery, and research while keeping confidentiality intact — sensitive terms minimized before dispatch, only zero-retention models in play, matter-level scoping, and a per-request log that answers a client, court, or regulator without an internal investigation.

What legal leaders are being asked

  • Which AI tools has this matter touched, and under what terms?
  • Was client material used to train a third-party model?
  • Can you demonstrate compliance with outside counsel guidelines on AI?
  • Who inside the firm accessed which model, and when?
  • What is your position under the EU AI Act for the systems you rely on?

These are evidence questions. They are unanswerable when access is a collection of individual accounts and answerable in minutes when every request passes one governed path.

Controls that matter for confidentiality

  • Zero Data Retention enforced per model, so no provider learns from client material.
  • Payload minimization for names, deal terms, and identifiers before dispatch.
  • Matter and client scoping, including ethical-wall restrictions on routing.
  • Attribution to individual timekeepers for disclosure and internal review.
  • Retention and legal-hold behavior you control, rather than a vendor's default.

Request an AI Readiness Review — free

A 20-minute review of how your teams use AI today, which models are in play, and the controls a board would expect. No obligation.

Frequently asked questions

Does using AI on client material waive privilege?

Privilege turns on confidentiality being maintained. Sending material to a provider that retains and trains on it undermines that argument; sending it under zero-retention terms through a controlled path with an audit record supports it. Document the control, not just the intent.

Can we use AI under outside counsel guidelines?

Increasingly clients permit it with conditions: named models, no training on their data, disclosure on request, and retained records. Those conditions are exactly what a gateway can enforce and evidence per matter.

How do we handle matter-level segregation?

Requests carry identity and matter context, so routing and logging can be scoped per matter and per client, including ethical-wall restrictions.

What about court and regulator disclosure requirements?

You need to state which tool, which model, and what was submitted. A per-request log answers that in minutes rather than through interviews.