Your AI Prompts Are Records. Does Your Retention Policy Cover Them?

When a federal court ruled earlier this year that a defendant’s AI chat history wasn’t privileged, coverage quickly settled on a single sentence: using AI waives privilege. Last week, Legal IT Insider’s lead analyst Neil Cameron revisited the ruling and arrived at a narrower, harder-to-act-on reading. The court never held that AI use waives privilege. It held that those particular prompts and outputs were documents — and documents get collected.

That distinction matters more to firm operations than the headline did.

What the ruling actually turned on

In United States v. Heppner, No. 25 Cr. 503 (JSR), Judge Jed Rakoff of the Southern District of New York found that roughly 31 documents recording a defendant’s exchanges with a consumer AI assistant were protected by neither attorney-client privilege nor the work product doctrine. The defendant used the tool himself, after receiving a grand jury subpoena, without direction from his lawyers. “Because Claude is not an attorney,” the February 17 written opinion reads, “that alone disposes of Heppner’s claim of privilege.” On work product, the court found the documents “were not prepared at the behest of counsel and did not disclose counsel’s strategy.”

Katherine Forrest, a former judge in the same district now co-chairing Paul Weiss’s global AI group, told Legal IT Insider the opinion was narrow partly because the briefing was thin, and that Rakoff did not set out to establish a standard for every prompt and output. Read that way, the ruling is less a verdict on AI than a reminder of something records managers have always known. If a system keeps a copy, someone can ask for the copy.

Which raises the question Cameron puts to firms directly: do you know whether your own tools retain prompts and outputs, and where those records sit?

For many firms, the honest answer is not yet, through no fault of anyone’s planning. The gap opened quietly while AI features arrived inside platforms the firm already owned, following the same pattern we wrote about when AI started showing up through existing systems, and the same question we raised about where an AI notetaker’s record goes after the meeting ends.

The records are already in the mailbox

Microsoft gave that gap new surface area last week. On October 15, it announced Copilot for legal, with more than 20 legal-specific skills generally available worldwide, document review and redlining inside Word, multi-file work in Cowork, and partner plugins connecting to research, document management, e-signature and e-discovery systems. Microsoft states that Copilot “processes prompts and responses within the Microsoft 365 service boundary” and does not use them to train its foundation models. Both points are worth having.

Neither point addresses retention, and Microsoft’s compliance documentation does. “Data from generative AI messages is stored in a hidden folder in the mailbox of the user who runs the AI app,” the Purview guidance on retention for Copilot and AI apps explains. Those messages, “including the hidden folders, are searchable by eDiscovery tools.” The same page adds a caution worth reading twice: “Messages visible in your AI apps are not an accurate reflection of whether they are retained or permanently deleted for compliance requirements.”

A configuration wrinkle lies beneath it. Copilot interactions were once swept into a retention location shared with Teams chats; Microsoft has since separated them. A policy written before that change may not reach what a firm assumes it reaches, and the only way to know is to look. Mailboxes already concentrate a firm’s correspondence, which is exactly why we spent last week on mailbox access control. AI interaction data now lands there, too.

Why this lands on the administrator’s desk

A retention question looks like a practice question until somebody has to answer it. Then it becomes a configuration question, a licensing question, and a documentation question — three things that tend to arrive in the firm administrator’s inbox regardless of who owns the policy on paper. Nobody should have to assemble that answer under deadline, during a collection, with outside counsel waiting.

So we assemble it in advance.

Innovative Computing Systems inventories where AI actually runs in a firm: licensed products, features enabled inside platforms the firm already pays for, and the plugins and connectors attached to them. We review the retention configuration for each of those surfaces against the firm’s records policy, so we document what the firm keeps and how long it keeps it rather than assuming it. We confirm status through the compliance tooling rather than an application’s visible history, because Microsoft’s own guidance says history isn’t the compliance record. This is the governance side of Managed Intelligence — acceptable use, approved tools, and pilot oversight handled together, so adoption and recordkeeping move at the same pace rather than one chasing the other.

Consumer-tier AI belongs in the same conversation. The Heppner record turned in part on a consumer privacy policy, and deciding which tools your people can use is a confidentiality control, not just a licensing preference. We handle that configuration and give attorneys an approved path that does the same work, so the control doesn’t become a workaround. It pairs with the verification habits we described, who checks the citations, and the measurement question clients are starting to ask.

You deserve a clear answer

Firms should not have to discover their retention posture during a dispute. For 37 years, since 1989, Innovative Computing Systems has worked exclusively with law firms and legal departments, and the job has not changed: build environments that are safe, secure, and reliable, and take ownership of the details that would otherwise land on your desk at the worst possible moment. We track where your AI records live, how long they stay, and who can search them.

To learn what your firm currently retains, request a consultation. A Solutions Consultant will follow up to schedule a conversation about your environment and walk through what your AI tools are keeping today.