A business owner wants a second opinion before terminating an important supplier. His lawyer has explained the contract’s weaknesses, the likely cost of litigation, and the company’s settlement options. An AI chatbot could organize that advice into a decision chart in seconds.
The owner uploads the email and asks: “What is our best move?”
The immediate result may be useful. The less obvious result is that the company has transmitted its lawyer’s analysis to an outside technology provider. Whether that information remains legally protected requires a closer look at the tool, its terms, and the circumstances of its use.
For businesses using AI, the decision about what to upload deserves as much attention as the answer the system produces.
Attorney-client privilege generally protects confidential communications between lawyers and clients provided as part of legal advice. A conversation with a chatbot does not become an attorney-client communication simply because the question concerns a legal problem. Uploading an existing lawyer-client communication raises an additional issue: whether sharing it compromises protection that already existed.
Work-product protection is a separate doctrine. It can protect materials prepared in anticipation of litigation by or for a party or its representative. It is qualified, and its rules differ from attorney-client privilege. Those differences help explain why courts have reached different conclusions about AI-assisted litigation materials.
In United States v. Heppner, a federal judge in New York addressed a criminal defendant’s use of Anthropic’s Claude to develop potential defense arguments. The defendant acted without counsel’s direction and later shared the resulting materials with his lawyers. United States v. Heppner, No. 25 Cr. 503, ECF No. 27 (S.D.N.Y. Feb. 17, 2026).
In its February 17, 2026 memorandum, the court held that the AI exchanges were protected by neither attorney-client privilege nor the work-product doctrine. Its reasoning included the absence of an attorney-client relationship with the chatbot, the confidentiality implications of the platform terms before it, and the lack of attorney direction or reflection of counsel’s strategy. Later delivery to counsel did not make previously unprotected materials privileged.
The decision is a warning against assuming that a personal AI account functions as an extension of the lawyer’s office. It does not establish that every AI product operates under identical terms.
Another decision illustrates why the analysis cannot stop there. In Morgan v. V2X, Inc., a Colorado federal magistrate judge recognized in a March 30, 2026, decision that a self-represented civil litigant could assert work-product protection for AI-assisted litigation preparation. The Court rejected automatic waiver based on AI use. Morgan v. V2X, Inc., No. 25-cv-01991-SKC-MDB, ECF No. 65 (D. Colo. Mar. 30, 2026).
These are federal trial-court decisions from outside New Jersey, not binding New Jersey precedent. They involved different procedural settings and different relationships between the litigant and counsel. However, these decisions show why businesses should examine the particular use before entrusting sensitive material to an AI system.
A practical starting rule is to require counsel’s approval before employees upload legal advice or litigation materials into an AI tool. That approval should address the specific task and account, including the contract and settings that govern it. A company subscription or an “enterprise” label should prompt review of the protections provided.
The review should answer concrete questions. Can the provider use the material to train or improve its models? Who may review it? How long are inputs and outputs retained? Can an employee generate a public sharing link? Do connected applications receive the information? What happens when the account is closed? Disabling model training addresses one concern; it does not, by itself, establish privilege or resolve every confidentiality issue.
The material also matters. A request to explain a publicly available contract term presents a different exposure from uploading counsel’s assessment of a witness or the company’s maximum settlement authority. Removing names may reduce identifying information while leaving sensitive facts and legal strategy intact. Counsel should decide whether a more general question can accomplish the business objective without submitting those details.
Information obtained in litigation requires particular care. Before using AI to summarize an opponent’s documents, review the protective order and any restrictions on disclosure. A business cannot assume that access to documents includes permission to transmit them to another service. Morgan shows that courts may impose specific conditions on that use.
Businesses should also plan for the records AI creates. A prompt may contain a manager’s account of events, assumptions about the dispute, or proposed negotiating positions. An output may combine those facts with errors or unsupported conclusions. Keeping the underlying records and distinguishing verified facts from generated text can prevent confusion when counsel later evaluates the evidence.
Relevant prompts and outputs may become subjects of discovery, subject to the ordinary limits of relevance, proportionality, privilege, and work-product protection. Once litigation is reasonably anticipated, preservation decisions should include relevant AI records.
If sensitive material has already been uploaded, tell counsel promptly. Identify the service, account type, upload date, information submitted, and any sharing that occurred. Preserve relevant records while counsel evaluates access restrictions, available remediation, and any applicable obligations. A quick, accurate account is more useful than an employee quietly deleting the chat and hoping the issue disappears.
AI can help businesses understand information and prepare for decisions. Its use should follow a considered process for handling legal material. Before uploading your lawyer’s email, agree with counsel on what may be shared, through which system, and for what purpose. That conversation is easier to have while the company still controls the information.
To find out more about our Business Litigation Group, click here.