Your AI assistant just read your client's confidential emails, drafted a response, and sent it. Did you waive attorney-client privilege?

Nobody knows. And that's a problem.

The Old Rules

Attorney-client privilege is simple in theory: communications between a lawyer and client, made for the purpose of seeking legal advice, are confidential and protected from disclosure.

Three requirements:

  1. Communication between attorney and client
  2. Made in confidence (no third parties present)
  3. For the purpose of legal advice

Breach any of these, privilege is gone. Forever.

Example: If you discuss your case in a crowded Starbucks, privilege is waived (third parties present). If you CC your buddy on an email to your lawyer, privilege is waived (not just attorney-client). If you forward legal advice to a business partner, privilege is waived (disclosure to third party).

The key word: third party.

And here's where AI breaks everything.

The Problem

Is your AI assistant a third party?

Courts haven't decided. Regulators haven't issued guidance. Bar associations are silent. We're in a gray zone, and lawyers (and clients) are making it up as we go.

The arguments:

AI = third party (privilege waived)

  • The AI model is operated by Anthropic/OpenAI/etc (third-party company)
  • Your confidential info is processed on their servers
  • Some providers use data for training (definite disclosure to third party)
  • Even with privacy commitments, data still leaves attorney-client control

AI = tool (privilege preserved)

  • It's like using spell-check or Westlaw search (nobody claims those waive privilege)
  • No human third party reads the data
  • Modern privacy policies exclude legal data from training
  • Attorney maintains control of output (review before sending)

The reality: It depends on the court, the jurisdiction, and how you configure your AI.

What We Know (So Far)

A few data points:

  1. ABA Formal Opinion 512 (2024): Lawyers may use AI tools if they take "reasonable measures" to protect client confidentiality. Didn't say whether AI itself is a "third party." Classic lawyer non-answer.

  2. California State Bar guidance (2025): Suggested that AI tools with strong privacy agreements (no training on user data, no human review) are "more like" tools than third-party services. Still not a firm rule.

  3. First federal case (2025, S.D.N.Y.): Judge questioned whether AI-drafted briefs preserved privilege when opposing counsel sought discovery of "all third-party tools used in document preparation." Case settled before ruling. We learned nothing.

So: no binding precedent, vague guidance, everyone guessing.

The Risk Scenarios

Where this gets messy:

Scenario 1: Litigation Discovery

  • You're in a lawsuit, opposing counsel demands all communications related to the case
  • You used Claude to draft internal strategy memos
  • Opponent argues: "Claude = third party, so those memos aren't privileged"
  • Court might agree (no precedent saying they're wrong)

Scenario 2: Inadvertent Disclosure

  • Your AI assistant auto-suggests a response based on prior client communications
  • You send it without reviewing
  • Recipient claims you disclosed privileged info (even if you didn't realize it)
  • Privilege might be waived for ALL related communications, not just that one email

Scenario 3: Regulatory Investigation

  • Government subpoenas your client communications
  • You claim privilege
  • They ask: "Did you use AI to draft these?"
  • If yes, they argue privilege doesn't apply (third-party involvement)
  • You might have to produce documents you thought were protected

Practical Guidance (For Now)

Until courts clarify, here's what lawyers should do:

1. Use AI Providers with Strong Privacy Terms

  • No training on your data
  • No human review of your inputs/outputs
  • Data encryption at rest and in transit
  • Ideally: on-premises or private deployment (if you have budget)

2. Disclose AI Use to Clients

  • Update engagement letters: "We may use AI tools in providing legal services. We take measures to protect confidentiality, but privilege rules are unclear."
  • Get informed consent
  • Document what AI you're using and how

3. Treat AI as a "Workflow Tool," Not a Decision-Maker

  • Use AI for drafting, research, brainstorming—but YOU review everything
  • Don't let AI make final legal judgments
  • Maintain attorney control over all client communications

4. Assume Worst-Case in Sensitive Matters

  • High-stakes litigation? Don't use AI on privileged communications.
  • M&A negotiations with strict confidentiality? Keep AI out of it.
  • Routine contract review? Probably fine.

5. Log Everything

  • Keep records of what AI tools you used, when, and for what purpose
  • If privilege is challenged later, you can show "reasonable measures" to protect confidentiality

What Clients Should Know

If you're hiring a lawyer, ask:

  1. Do you use AI tools?
    (If they say no, they may not be up to date on current industry practice.)

  2. Which providers, and what are their privacy policies?
    (If they don't know, find a different lawyer.)

  3. How do you protect privilege when using AI?
    (If they shrug, RUN.)

The Uncomfortable Truth

The legal profession is flying blind. We're using AI because it's too useful not to. But we're doing it without clear rules, and clients are exposed.

My prediction: Within 2 years, we'll see a major privilege fight over AI use. A court will rule one way or the other, and half the profession will realize they've been doing it wrong.

Until then? Be paranoid, document everything, and assume opposing counsel is reading the same AI ethics guidance you are (which is to say: not much).