AI Conversations Are Records

Short Answer

Prompts, outputs, uploaded files and logs can be electronically stored information. Relevant records must be preserved when the governing litigation standard is met, and discovery follows relevance, proportionality and protection rules. In the consolidated OpenAI copyright litigation, an October 9, 2025 order (filed as Document 922 in The New York Times Company v. Microsoft Corporation, No. 23-cv-11195 (S.D.N.Y.), captioned In re OpenAI, Inc., Copyright Infringement Litigation, No. 25-md-03143) terminated the broad forward-looking duty to preserve and segregate output logs effective September 26, 2025, while continuing preservation of previously segregated logs (except requests originating in the European Economic Area, Switzerland, or the United Kingdom), continuing preservation of logs from specified accounts, and leaving ordinary Rule 37(e) duties unchanged; separate production orders addressed a retained sample of about 20 million de-identified logs. That case illustrates both the reach and limits of discovery. A business needs a records policy that identifies AI accounts and vendor copies, plus a hold process that can actually reach them.

Which Laws Apply

Texas AI-specific: none for private businesses; public entities, see Government AI and Public Records.

Generally applicable Texas law: Texas Rules of Civil Procedure on discovery of electronic data (Tex. R. Civ. P. 196.4); Texas spoliation law (Brookshire Brothers v. Aldridge, Tex. 2014); Public Information Act for governmental bodies.

Federal: Fed. R. Civ. P. 26, Fed. R. Civ. P. 34 and Fed. R. Civ. P. 37(e).

Why AI Records Are Different From Email

AI use can create candid prompts about legal exposure, successive versions of a document and copies distributed across vendor logs and backups. A records inventory should account for those copies and the terms controlling them. The same workflow that makes retrieval useful can leave evidence of the inputs, the changes and the decision.

The OpenAI Preservation and Production Orders

A May 13, 2025 order in the OpenAI copyright litigation required preservation and segregation of specified output logs that would otherwise be deleted. The underlying suit is The New York Times Company v. Microsoft Corporation, No. 23-cv-11195 (S.D.N.Y.), in which OpenAI is a party; copyright suits against OpenAI are consolidated as In re OpenAI, Inc., Copyright Infringement Litigation, No. 25-md-03143 (SHS) (OTW). An October 9, 2025 stipulation and order, filed as Document 922 on the NYT docket and captioned to the MDL, terminated OpenAI’s broad forward-looking obligation to preserve and segregate output log data effective September 26, 2025 (paragraph 1). Previously segregated logs remained subject to preservation, except logs for requests originating in the European Economic Area, Switzerland, or the United Kingdom (paragraph 2); OpenAI also agreed to preserve, on a going-forward basis, logs from accounts tied to domains the News Plaintiffs identified, regardless of location (paragraph 3). The order did not change ordinary Federal Rule of Civil Procedure 37(e) duties (paragraph 4). Separate November 7, 2025 production and December 2, 2025 reconsideration orders in the MDL concerned a retained sample of about 20 million de-identified logs. The December 2 order, MDL ECF 812, denied reconsideration under case-specific relevance, proportionality and protective-order analysis. Preservation and production are different obligations; neither is a nationwide rule requiring every chatbot provider or user to retain everything.

Those orders bound the AI provider, but the reasoning applies to any business that holds AI records relevant to a dispute.

Preservation Duties

Under Texas law, a party must preserve evidence when it knows or reasonably should know that there is a substantial chance a claim will be filed and that the evidence is relevant; the trial court may remedy spoliation, including through instructions in serious cases. In federal court, Fed. R. Civ. P. 37(e) governs loss of electronically stored information that should have been preserved. A hold notice that lists email and documents but not AI tools leaves a gap. Hold notices should name enterprise AI tools, the vendor’s retention settings and any personal accounts employees use for work.

Discovery

A discovery request can reach AI prompts and outputs if they fall within its terms and governing relevance and proportionality limits. Privacy is part of that analysis, not a categorical answer either way. The OpenAI court ordered specified production under a protective regime; it did not abolish privacy objections in other disputes. Identify the requested account, period, data and alleged relevance, then determine whether filtering, redaction or a protective order is appropriate.

What to Decide Now

Which AI tools are approved for work and which accounts they use.

Retention periods for prompts, outputs and logs, set to business need and legal duty.

How a litigation hold suspends deletion in each tool, including at the vendor.

Whether employees may use personal AI accounts for work; if they do, those records may still be discoverable.

Training that tells employees what AI records are and that they may be read by others.

Real Examples

In re OpenAI (S.D.N.Y. 2025 to 2026): preservation and production of AI logs.

United States v. Heppner (S.D.N.Y. 2026): a defendant’s AI research was not privileged and was available to the government; see Privilege and AI Tools.

What Is Unsettled

How proportionality will limit requests for broad AI logs in ordinary commercial cases; whether courts will require production of AI records from personal accounts used for work.