Skip to main content
AI confidentiality & privilege

Efficiency is not worth a waiver. It never has to be one.

Confidentiality and privilege protection in AI-assisted work, from Aun & Co.: data flows designed so professional secrecy survives the tools that speed the work.

For lawyers, advisers and any profession built on secrecy, AI raises one question above the rest: does using these tools compromise confidentiality or privilege. The honest answer is — it depends entirely on the architecture. Client material entering a consumer tool under training-permissive terms is a genuine exposure; the same work routed through properly-termed systems, with classified inputs and human gates, is not. The firm designs that architecture, and practises it: its own privileged work runs through exactly the flows it builds for clients.

The work spans
  • Data-flow design: which document classes may touch which systems
  • Vendor-terms alignment with professional secrecy duties
  • Privilege analysis of AI-assisted work under Israeli law
  • Client-consent and engagement-letter language where appropriate
  • Breach-scenario mapping and the response posture for each
  • Your team uses AI on client work and no one has analysed what that does to privilege.
  • A client asked directly whether AI touches their matters and the answer must be precise.
  • Draft pleadings or advice pass through tools whose terms you have never mapped to your secrecy duties.
  • You want AI's speed on document-heavy work without exposing the protected core of the file.

The method is classification plus routing: every document and data class in the practice is tiered by sensitivity, and each tier is assigned the systems it may touch — from open tools for public material to no-AI zones for the most protected work. Vendor terms are matched to the tiers they serve, and the human gate is placed where output re-enters the professional product. The result is written as a one-page flow map the whole team can hold in mind.

04 · What you get

Tiered by sensitivity

A classification that lets routine material move fast while the protected core never touches an unvetted system — speed and secrecy by routing, not by choosing.

Terms matched to duties

Each tool's contractual reality mapped against professional-secrecy obligations, so the architecture rests on documents rather than assumption.

An answer for clients

A precise, honest account of how AI touches their matters — increasingly a question sophisticated clients ask before engagement.

A typical engagement: a law office adopting AI for drafting commissions a confidentiality architecture. Documents are tiered, two vetted tools are matched to the lower tiers, the privileged core is walled off, and the engagement letters gain a paragraph that answers the client question before it is asked.

Described in abbreviated, anonymised form to preserve client confidentiality.

Does using AI on client documents waive privilege?

Not inherently — privilege turns on confidentiality being maintained, and a properly designed flow maintains it: tools contractually barred from training on or disclosing inputs, restricted access, and professional control of the output. The risk is in unexamined flows, not in the category of technology.

Can confidential material ever go into public AI tools?

Identifiable client material should not enter tools whose terms permit training or human review of inputs — the standard posture of consumer tiers. Enterprise configurations with training exclusions and retention limits are a different contractual animal; the tier distinction is the whole game.

What does a confidentiality architecture engagement produce?

A document classification, a flow map assigning each tier its permitted systems, vendor-terms verification for each, and the client-facing language for engagement letters. For a small practice it typically takes two to three weeks — and it is the prerequisite the rest of AI adoption stands on.

Start a conversation.

The firm replies within one business day.