Protecting attorney-client privilege, work product, and confidential client information when using LLMs in legal practice
Attorneys have heightened ethical obligations to protect client confidences, preserve attorney-client privilege, and maintain competence in technology. Using LLMs with client information requires careful analysis of privilege waiver risks, confidentiality breaches, and compliance with professional conduct rules. Multiple state bars have issued opinions on AI usage, and the stakes for getting it wrong include malpractice liability, disciplinary actions, and waiver of privilege.
Disclosing privileged communications to third parties can waive attorney-client privilege and work product protection. If client communications sent to LLM vendors are deemed disclosures to third parties without adequate safeguards, privilege may be lost - potentially exposing sensitive strategy, advice, and communications in litigation.
Two separate protections exist for legal communications and materials:
Purpose: Protects confidential communications between attorney and client for the purpose of seeking or providing legal advice.
Requirements:
Examples: Client emails to attorney asking legal questions, attorney advice memos to client, communications about litigation strategy.
Purpose: Protects materials prepared in anticipation of litigation from discovery (broader than privilege).
Two Tiers:
Examples: Witness interview notes, litigation research memos, trial strategy outlines, attorney's thoughts on case strengths/weaknesses.
Voluntary disclosure to a third party generally waives attorney-client privilege. Key questions for LLM usage:
The ABA Model Rules of Professional Conduct govern attorney behavior. Most states have adopted these rules (with variations):
Requirement: "A lawyer shall provide competent representation to a client. Competent representation requires the legal knowledge, skill, thoroughness and preparation reasonably necessary for the representation."
Comment 8 (2012 Amendment): Competence includes "keep[ing] abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology."
LLM Implication: Lawyers must understand how LLMs work, their limitations (hallucinations, outdated information), and risks (privilege waiver, data exposure). Simply using AI without understanding it may violate competence duties.
Requirement: "A lawyer shall not reveal information relating to the representation of a client unless the client gives informed consent..."
Comment 18: Discusses lawyers' duty to "act competently to safeguard information relating to the representation" against unauthorized access.
LLM Implication: Sending client information to LLM vendors may constitute "revealing" information unless: (1) client gives informed consent, OR (2) vendor is considered an agent with confidentiality obligations, OR (3) information is anonymized/de-identified sufficiently.
Requirement: "A lawyer shall make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of a client."
LLM Implication: Reasonable efforts include: (1) vetting LLM vendor security (SOC 2, encryption), (2) using zero-retention configurations, (3) training staff on what data can/cannot be sent to LLMs, (4) monitoring for breaches.
Requirement: Lawyers must ensure that nonlawyers (paralegals, vendors) act compatibly with the lawyer's professional obligations.
LLM Implication: If LLM vendor is considered a "nonlawyer assistant," lawyer must ensure vendor complies with confidentiality and privilege obligations. This requires contractual protections and due diligence.
Multiple state bars have issued ethics opinions on AI usage. While not binding in all jurisdictions, they provide guidance:
1. Competence Required
Lawyers must understand AI tools sufficiently to evaluate their reliability, limitations, and risks. Blindly accepting AI outputs violates competence duties.
2. Verify AI Outputs
Lawyers must independently verify legal research, citations, and factual claims generated by AI. Several high-profile cases involved sanctions for citing fake cases from ChatGPT.
3. Confidentiality Safeguards
Before inputting client information into AI tools, lawyers must: (a) review vendor terms of service, (b) ensure data isn't used for training, (c) assess data security, (d) consider client consent.
4. Disclosure to Clients
Some opinions suggest lawyers should inform clients if AI is used for substantive work (not settled law, but prudent practice).
5. Billing Considerations
Billing for AI-generated work must be reasonable. Cannot bill full associate hours for work completed primarily by AI with minimal review.
Before using an LLM service with client information, law firms should evaluate vendors on these criteria:
Essential contract terms:
Required security measures:
Red flags to watch for in consumer/free-tier services:
Ensure LLM usage doesn't waive privilege:
These enterprise-tier services offer contractual protections suitable for law firms:
These services should NOT be used with client confidences or privileged information:
LLMs present unique inadvertent disclosure risks that lawyers must manage:
When uploading documents to LLMs for analysis (PDFs, Word docs), metadata can include:
Solution: Strip metadata before uploading documents to LLMs, or use text-only extracts.
Some LLM interfaces store conversation history accessible to other users with shared accounts or through search features:
Solution: Use enterprise accounts with separate user access; disable chat history; use API access instead of web interface.
If LLM vendor uses your data for training, privileged information could later appear in responses to other users:
Solution: Only use vendors with explicit "no training" guarantees; verify in contract.
Law firms can safely use LLMs for many tasks, with appropriate safeguards:
Researching legal principles, statutes, and case law on general topics (not case-specific).
Safeguard: Verify all citations independently; don't reveal client facts
Generate initial drafts of common documents (NDAs, engagement letters, generic motions).
Safeguard: Use anonymized facts; remove all client identifiers before inputting
Generate outlines or question lists for depositions and briefs based on public legal principles.
Safeguard: Don't include case-specific facts, witness names, or confidential strategy
Draft initial versions of client emails or letters on routine matters.
Safeguard: Use vendor with no-training guarantee; review for accuracy and tone
Analyze publicly available contracts to identify standard clauses and market terms.
Safeguard: Only use public contracts; not confidential client agreements
Create training modules, CLEs, or practice group updates on legal developments.
Safeguard: Use public information only; no client examples without anonymization
These use cases should ONLY be done with legal-specific AI tools designed to preserve privilege:
We can help you navigate attorney-client privilege, confidentiality obligations, and ethical rules to safely leverage LLMs while protecting your clients.
Schedule a Consultation