Legal Holds for AI Assistants: Preserving Chats Before They Age Out

Most AI assistants are built to forget. Deleted conversations are often removed on short timers, and in 2025 that design choice collided with litigation. In The New York Times' copyright case, a court order required OpenAI to retain deleted ChatGPT conversations and API content that would typically be removed from its systems within 30 days, as OpenAI described in its public response to the preservation demand. The obligation fell on the AI provider rather than on the organizations using its tools, but the lesson carries over to every enterprise: when a duty to preserve attaches, retention timers do not pause on their own.

Scale makes the issue urgent. McKinsey's State of AI in 2025 survey found that 88% of organizations now use AI in at least one business function, up from 78% the year before. Each of those deployments produces prompts, responses, and summaries that may become relevant to a dispute, an internal investigation, or a regulatory inquiry.

AI Chat Data Has a Shorter Shelf Life Than Email

Email and documents typically sit under retention policies measured in years. AI assistant content often follows different rules:

• Chat histories may be deleted automatically after a set number of days

• Users can delete individual conversations, sometimes without administrators being notified

• Temporary or private modes may never store content in a reviewable form

• AI-generated summaries, drafts, and meeting notes may be saved in collaboration platforms rather than in the assistant itself

These differences mean a hold notice alone may not stop content from disappearing. Any AI chat retention legal hold decision needs to account for where content lives and how long it survives by default. For a broader view of this data category, see this overview of legal holds for AI-generated content.

Preservation Starts With Knowing Where AI Content Lives

Map the AI Tools in Use

Inventory sanctioned assistants, AI features embedded in collaboration suites, and tools adopted at the team level. Record who administers each tool and which retention settings apply.

Identify Every Storage Location

A single interaction can leave records in several places: the assistant's chat history, a generated document in shared storage, a message posted to a channel, or a transcript attached to a meeting. This guide on how to preserve AI-generated content in collaboration platforms explains how those artifacts are distributed.

Document Default Retention Windows

Record how long each tool keeps prompts and responses, whether deletion is user-controlled, and whether administrators can override it. This record supports defensibility if preservation decisions are questioned later.

A Defensible Legal Hold Process Covers Four Steps

1. Trigger the Hold Early

The duty to preserve generally begins when litigation is reasonably anticipated. For AI content with retention windows of 30 days or less, a delay of a few weeks can mean permanent loss.

2. Scope Custodians and AI Data Sources Together

Hold scopes should name the AI tools each custodian uses, not only mailboxes and file shares. Include prompts, responses, attachments, and metadata such as timestamps and user identifiers.

3. Apply Technical Preservation, Not Only Notices

Custodian notices remain necessary, but they cannot stop automated deletion. Pair notices with system-level holds or in-place preservation that suspends retention policies for in-scope content.

4. Verify, Monitor, and Release

Confirm that preservation is active for each source, track new AI tools as they are adopted, and release holds formally when the matter closes so normal retention resumes.

The comprehensive guide to legal holds outlines how these steps fit into the broader lifecycle of legal holds.

Common Gaps Put AI Content at Risk

• Relying on user compliance to keep conversations that users can delete themselves

• Overlooking embedded AI features that save output in collaboration tools rather than in a standalone assistant

• Collecting text without context, such as the prompt that produced a response or the thread around it

• Treating AI holds as a one-time setup instead of reviewing scope as tools and settings change

Each of these gaps is easier to close before a matter begins than after a preservation demand arrives.

Connected Platforms Make AI Holds Repeatable

Manual preservation across many AI tools and collaboration platforms is difficult to sustain across matters. A platform that connects directly to these sources can apply holds, preserve content with its metadata, and maintain an audit trail from one place. Onna and Reveal Hold connect hold issuance to in-place preservation across collaboration and cloud data sources so that legal, compliance, and IT teams can preserve and collect content consistently. Teams evaluating their options can request a demo to see how AI-generated content fits into an existing legal hold workflow.

Preserving AI Chats Is a Timing Problem First

Organizations that are well prepared for AI-era discovery treat AI assistants as standard data sources from the first day of a matter. Mapping where AI content lives, knowing how long it survives, and applying technical holds early closes the window in which relevant chats can age out.

To build a legal hold approach that preserves AI-generated content before retention settings remove it, contact the Onna team.