Skip to content
Judicial AI Standards Institute sealJudicial AI Standards Institute

The Bench & Bar AI Brief | Issue 005 | June 23, 2026

AI Work Now Leaves a Record

New court and bar signals show the next legal AI duty: manage the record AI creates, including chats, prompts, copied materials, citations, competence training, and privilege claims.

Back to BriefOpen checklist

Tennessee Edition

Welcome to Issue 005 of The Bench & Bar AI Brief, Tennessee Edition.

The first legal AI lesson was simple: do not file fake cases.

The next lesson is broader: AI work leaves a record.

That record may include prompts, chat logs, copied documents, generated text, citation lists, model names, retention settings, and the human decisions made after the tool answered. Courts and lawyers now have to ask the same practical questions they ask about email, drafts, expert notes, research files, and litigation holds: what was created, who saw it, where did it go, and must it be preserved or produced?

The Texas Business Court item is the cleanest signal this week. Legal sources report that in Tate Group Automotive, LLC v. Legacy Automotive Capital, LLC, Judge Grant Dorfman reviewed a non-lawyer principal's AI conversations in camera and treated some AI chats prepared in anticipation of litigation as work product under Texas procedure. The same reports also describe limits: materials or products shared with ChatGPT may still need to be disclosed. Judicial AI Standard did not retrieve the primary minute entry, so this issue uses the source reports narrowly and does not quote the order.

The Connecticut Shell Oil discovery fight points in the same direction from another angle. Public filings confirm that an order concerned artificial-intelligence prompts or queries used by an expert team in preparing a report, and those filings also confirm the order was challenged and stayed pending objections. That posture matters. The lesson is not that every prompt is automatically discoverable. The lesson is that AI prompts can become litigation evidence, privilege evidence, or waiver evidence.

The sanctions line is still moving too. In Withers v. City of Aberdeen, a Northern District of Mississippi order addressed hallucinatory citations and imposed serious lawyer consequences. In Joyce Barber v. Lawrence J. Morawa, MD, the Michigan Court of Appeals addressed AI-generated fabricated authority and remanded for a sanctions determination against counsel personally. Those are not anti-technology opinions. They are professional-duty opinions: verify the record before asking a court to rely on it.

The competence signal is just as important. Florida Bar News reports that judges at The Florida Bar Convention told lawyers that AI competence is no longer optional. That is not Tennessee authority, but it is a useful bench-and-bar warning. AI literacy is becoming part of supervision, candor, confidentiality, court administration, and ordinary legal competence.

A Tennessee bench-and-bar checklist.

  1. 1Treat AI chats and prompts as possible litigation records.
  2. 2Decide before tool use what may be privileged, confidential, work product, discoverable, or barred from the tool entirely.
  3. 3Do not paste protected facts, sealed material, or sensitive client information into unmanaged systems.
  4. 4Preserve prompts, chats, outputs, and model details when the work may later be challenged.
  5. 5Verify every case, quotation, statute, rule, and procedural claim outside the AI tool.
  6. 6Train lawyers, staff, clerks, and vendors on what the tool can do, what it cannot do, and when human review is mandatory.

Bench & Bar takeaway.

The good rule is not panic. It is record discipline. If AI helps produce legal work, the human user still owns the legal duty, and the institution still owns the governance problem.

The question for this week is simple: if someone asked tomorrow how an AI-assisted filing, order draft, expert report, or agency decision was made, could your office answer from a reliable record?

Three questions for local discussion.

  1. 1If someone asked tomorrow how an AI-assisted filing, order draft, expert report, or agency decision was made, could your office answer from a reliable record?
  2. 2What AI prompts, chats, outputs, and model details should be preserved when work may later be challenged?
  3. 3What should be barred from unmanaged AI tools before confidentiality, privilege, work-product, or sealed-record problems arise?

Distribution

Keep up between issues.

Subscribe to the weekly Brief by email, and follow Judicial AI Standard on X for daily AI and courts updates when verified information is available.

Subscribe to the BriefFollow on X
Return to The Bench & Bar AI Brief archive