THE SIGNAL IN ONE SENTENCE

A conversation with an AI can feel confidential without automatically receiving the legal protection attached to work prepared by a lawyer or for litigation.

01

WHAT ACTUALLY CHANGED

A September 7 legal analysis collected the first conflicting court decisions about AI-assisted litigation preparation. The cases do not create one national rule. They show courts beginning to ask whether chatbot prompts and outputs were prepared because of litigation and by someone covered by the work-product doctrine.

Federal courts in Michigan and Colorado allowed self-represented litigants to claim work-product protection over some ChatGPT materials. A New York court also quashed a subpoena seeking a litigant’s records from OpenAI. Those decisions recognized that a person without counsel can still prepare material in anticipation of litigation.

A Massachusetts court reached a different result when a represented plaintiff’s girlfriend created the AI materials. The court concluded that she was neither the party nor a qualifying representative acting under counsel’s direction, so the materials did not receive the asserted protection.

A particularly common situation remains unsettled: a represented client independently uploads facts or documents about an active case before discussing the work with a lawyer. Protection may turn on jurisdiction, purpose, direction, product settings, disclosure to others, and the exact record requested in discovery.

02

WHY THIS MATTERS

People increasingly use chatbots as the first room where they think through a frightening problem. The interface feels private and conversational. The law does not award confidentiality based on the emotional texture of the box.

Attorney-client privilege and work-product protection are different doctrines with specific requirements. A chat may contain sensitive facts without qualifying for either. It may also be retained, exported, subpoenaed, or disclosed in ways the user did not anticipate.

Lawyers and clients need a plain rule before the prompt is typed. Decide which tools are approved, what information may enter them, whether counsel is directing the work, how records are retained, and how AI material will be handled in discovery. The worst time to invent that policy is after opposing counsel asks for the transcript.

FIG. 060CONFIDENTIAL FEELING IS NOT LEGAL STATUS
1PERSON PROMPTS→
2AI STORES RECORD→
3LITIGATION PURPOSE→
4COURT TESTS PROTECTION→
5DISCLOSE OR WITHHOLD
Who created the material, under whose direction, for what purpose, and in which court can determine whether the conversation remains protected.

03

WHERE IT COULD HELP

  • Set approved AI tools in client engagement policies
  • Warn clients before they upload case documents
  • Keep attorney direction connected to research tasks
  • Address AI records in discovery and protective orders

KEEP A HAND ON THE WHEEL

This article reports developing decisions and is not legal advice. Rules vary by jurisdiction and facts. Anyone facing litigation should ask qualified counsel before putting confidential, privileged, identifying, or strategically sensitive information into a general-purpose AI service.

04

TERMS WORTH KEEPING

SOURCES AND VERIFICATION STATUS

This article was written from the materials below. Product claims and dates were checked against those sources on September 7, 2026.

PUBLICATION RECEIPT: Revision 1. Approved by Zak and published September 7, 2026.

THE PUBLICATION ENGINE

WANT A SIGNAL OF YOUR OWN?

We build source-grounded publications, private briefings, and editorial systems for organizations with something useful to say.

WORK WITH US