For executives and their teams

Use your favorite AI on your legal questions & keep it confidential.

Everything your team pastes into ChatGPT or Claude is discoverable in a lawsuit. Privilege AI is a legal chat workspace with your attorney in the session, so your questions stay confidential and counsel has the record to claim privilege over the conversation.

Acme Inc. · Vendor NDA · Privileged
Marketing Team Lead

Can I sign this vendor NDA, or does the non-solicit clause overreach?

Claude Sonnet 4.6

Standard on confidentiality Doc 2 · p.1 , but §6 adds a two-year non-solicit covering every employee Doc 2 · p.3 .

In-house counsel

Does §6 survive termination, and is two years enforceable in New York?

Claude Sonnet 4.6

It survives two years past the end date Doc 2 · p.4 . New York upholds that only to protect a legitimate interest — narrow it to the engagement team first.

Privileged session · Counsel presentRouted: Claude Sonnet 4.6
Every message logged · tamper-evident record · 09:24:11 EDT

Every one of these, asked in a consumer chatbot, is a discoverable business record. Asked here, it’s protected.

Privilege AI works with the following.

AnthropicClaude Fable 5Claude Opus 4.8Claude Sonnet 5Claude Sonnet 4.6
OpenAIGPT-5.5GPT-5.4GPT-5.4 mini
The problem

Private is not privileged.

Your people are not going to stop. Contracts, NDAs, “can we do this?” — that work already runs through ChatGPT and Claude, and an enterprise plan does keep it private: encrypted, access-controlled, excluded from training.

But private only means nobody else can read it today. Privileged means the other side cannot make you hand it over. A chat with no attorney in it — however private — is an ordinary business record: in a lawsuit or an investigation it can be demanded, produced, and quoted.

Privilege AI changes what that record is. Your attorney is in the session, the conversation is tied to a legal matter, and every exchange is logged with who said what and why — so counsel has what they need to claim the conversation as privileged, and to defend the claim.

Privilege AI

One workspace for every AI session that has to stay privileged. Every model. Every matter.

Privilege AI is the workspace where you and your attorney put AI to work on the same question, in the same session. Every conversation belongs to a legal matter, and every answer is recorded with its citations, its model, and who was present — the record that lets counsel support a claim of privilege. Bring your company’s own OpenAI and Anthropic keys, or use a system key.

Your attorney is in the session

You and in-house counsel prompt the same AI on one shared transcript, and counsel reviews the basis — the co-working is what supports the claim of attorney-client privilege.

Bring the model. Own the record.

Connect your own ChatGPT Enterprise or Claude for Work keys. Or use a system key. Your sessions never train anyone else's model, and the record stays yours.

Built around the matter

Sessions live inside a matter, not a chat window. Documents, work product, privilege log, and audit ledger are scoped to the matter and travel together.

Who it's for

Built for the people who ask.
Bought by the lawyers who answer.

For executives and their teams

Use the best AI on real legal questions — and keep it privileged.

Review a contract before you sign it, ask what an NDA actually commits you to, or work through an employment question — with your in-house attorney in the same session. Confidential here goes further than private: your lawyer is present, so counsel can claim privilege over the conversation and has the record to defend it.

  • Review contracts, NDAs, and employment questions inside a matter, not in a consumer chatbot.
  • Your attorney prompts the same AI alongside you and reviews every answer — that involvement is what supports the privilege claim.
  • Answers cite the documents in your matter, so you can see where the advice comes from.
  • Single sign-on, and a sign-out-everywhere control if a device is lost.
Capabilities

What the product actually does.

Chat across the leading models, with counsel in the session.

Use OpenAI or Anthropic. Your attorney works in the same matter, and answers cite the documents you uploaded.

See a sample session

The privilege basis is recorded for you.

Each turn logs why it is privileged (AC, WP, AC+WP) with the rationale, so counsel can assert and defend it later. Exports in Standard, Categorical, S.D.N.Y., and N.D. Cal. formats.

How the basis is logged

Documents you can point to.

Upload the contract or NDA, ask about it, and get answers that cite the page. Version history, a 30-day restore window, and a sensitive-content scan before anything is shared.

See cited answers

Your edits saved as work product.

Tighten a memo or a board update in the session; the new version is saved and searchable, not lost in a chat thread.

Inside the matter workspace

An audit chain you can stand behind.

Every message, edit, and export is hash-linked to the one before it. If the record is ever tampered with, a non-dismissible banner says so.

Read the trust profile

Counsel stays in control.

Your legal team sets who can see what, and break-glass access takes two approvers and lands a full record on the matter audit chain.

How access is controlled
Security and privilege

What keeps your session privileged and provable.

Privilege depends on showing the basis, the boundaries, and the chain. Privilege AI produces all three as a byproduct of normal use — and the controls below are what make that record hold up.

SOC 2 Type II and ISO 27001 audited.

Independently audited against SOC 2 Type II and certified to ISO 27001. The controls they cover — the ones on this page — are in production today; the reports are available under NDA on request.

Tenant isolation.

Postgres row-level security plus per-matter data-encryption keys, wrapped by a master key. Rotation is a one-click operation, logged.

Encrypted key storage.

Provider API keys are stored AES-256-GCM encrypted. Matter data keys are wrapped by a master key (libsodium). Customer-managed KMS is on the roadmap, not shipped yet.

No training on your matters.

Customer matter content is not used to train any model, ours or a provider's. The provider routes use the no-train endpoints.

Hash-linked audit chain.

Every action — message, edit, view, export — is SHA-256 chained to the previous. Verifiable from the UI.

Configurable retention and legal hold.

Per-matter retention windows. Legal hold suspends deletion. Deletion is name-typed with a 30-day restore window.

Read the full trust profile

The record

Every session produces the record that backs the privilege.

01

Your question

02

Matter scope + privilege basis

03

Counsel co-prompting the same AI in the session

04

Model route (provider, model, route reason)

05

Response with inline citations

06

Privilege log entry (auto-drafted, counsel-edited)

07

Audit chain entry (hash a3f1…7c2b)

Privilege AI does not assert privilege for you. It produces the basis, the citations, the route, the timestamps, and the chain — so counsel can.

Provisioned by legal teams.

AgencyAuditsuisse Assurance
In practice
“We rolled it out so our employees could use ChatGPT and Claude on contracts and HR questions without us losing privilege. The basis and the audit trail are just there.”

General Counsel, Fortune 500 enterprise — name available under NDA.

Privilege AI is provisioned by legal departments and firms so their people and clients can use AI on employment, contracts, NDAs, and other questions — with counsel in the loop. References available on request.

SOC 2
Type II · independently audited
ISO 27001
certified
SHA-256
hash-chained audit ledger
0
matters used to train any model
Pricing

Pricing is by seat, with volume discounts for legal teams.

Every seat gets matter workspaces with counsel in the session, routing to no-train endpoints on enterprise keys, privilege log exports, and the hash-chained audit ledger. Pricing scales with the people you provision and the matters they run.

Talk to sales
Questions

Frequently asked.

  • Do I need to be a lawyer to use this?

    No. Privilege AI is built for employees and executives. You ask the questions in plain terms and get answers that cite your documents; your in-house attorney is in the same session, prompting the same AI alongside you.

  • What kinds of questions do people ask?

    Reviewing a contract before you sign it, understanding what an NDA actually commits you to, and hypothetical employment or contract questions are the common ones — always inside a matter, with counsel in the session.

  • How does my attorney stay in the loop?

    Your attorney is added to the matter and prompts the same AI alongside you, on one shared transcript. They ask their own follow-up questions and review the privilege basis on each turn — that involvement is what supports the claim of privilege.

  • Is the conversation privileged just because I used Privilege AI?

    No. Privilege depends on the relationship, the purpose, and the confidentiality of the communication, and on your attorney being involved. Privilege AI produces the record — basis, citations, route, audit chain — so your counsel can assert and defend the privilege.

  • Who buys it — me or my legal team?

    Your in-house legal department or an outside firm provisions Privilege AI and adds you to a matter. You use it; they stay in the loop and keep the work-product protection.

  • Which models can I use?

    OpenAI GPT models and Anthropic Claude. Your legal team brings enterprise keys, or uses a system key. The model used on each message is part of the audit record.

  • Will my matter content be used to train any model?

    No. We route to provider no-train endpoints and we do not train our own models on customer matter content. This is contractually committed.

  • Can I use this with outside counsel?

    Yes. Outside counsel can be added as members on a matter; the outside-counsel toggle records the relationship for common-interest privilege purposes.

  • How is the audit chain verified?

    Each event row is SHA-256 chained to the previous. A verifier runs on read; if it ever breaks, a non-dismissible banner shows on every surface until it is reconciled.

  • Is Privilege AI SOC 2 certified?

    Yes. Privilege AI has completed a SOC 2 Type II audit and is certified to ISO 27001. The controls they cover — tenant isolation, encrypted key storage, the hash-chained audit ledger — are in production today, and the reports are available under NDA on request.

Bring AI into your matters. Keep the privilege.

Book a 30-minute walkthrough. We'll open a matter, run a session, and produce the privilege log and audit chain on screen.

Book a demoRead the architecture brief

Or email counsel@onprivilege.com with a memo on a matter you're trying to run.