Federal Courts Tighten Rules on AI Use in Discovery
Two federal court decisions are establishing how judges will police generative AI in litigation. Rather than settle the unsolved questions around privilege and waiver, Jeffries v. Harcros Chemicals, Inc. and Morgan v. V2X, Inc. show courts using protective orders to control how lawyers handle AI tools with discovery documents.
The courts didn't ban AI. They imposed restrictions because public AI tools create real risks: confidential data could be absorbed into model training, stored indefinitely, or exposed online.
Blanket Bans on Public AI Tools
In Jeffries, the Kansas federal court prohibited uploading any produced documents into public AI tools. The judge identified specific risks that justified the ban: data incorporated into model training may be impossible to retrieve, storage practices are unclear without contracts, and sensitive or regulated information could become public.
The court reasoned that the restriction actually encouraged fuller discovery. Without the ban, lawyers might under-produce documents or over-redact to protect against AI exposure. The ruling treats AI limits as tools for better disclosure, not obstacles to it.
Vendor Requirements Now Judicial Scrutiny
The Colorado federal court in Morgan went further, setting minimum standards for any AI tool used with confidential information. The tool must prohibit training on inputs, restrict third-party access, and allow deletion of data on request.
Those requirements eliminate most commercial "open" AI systems. Enterprise-level or custom-built tools become the only option. The court acknowledged this creates access problems for under-resourced litigants but concluded the confidentiality risk outweighed those concerns.
Because no binding precedent defines acceptable AI vendor safeguards, choosing a tool with flexibility to meet evolving court standards is critical. The wrong choice could bar a party from using AI entirely in discovery.
AI Tool Identity Is Discoverable
Courts are now treating AI use itself as a discovery issue. Morgan required a party to disclose which AI tool it used with confidential materials, though the court protected the substantive legal analysis behind the tool selection.
The fact of AI use is discoverable. The strategy behind it remains protected. Disputes will likely determine how much disclosure is necessary to address confidentiality concerns without piercing work product protection.
Plan Early or Face Consequences
Both decisions underscore that lawyers should raise AI issues at Rule 26(f) conferences and build protections into protective orders from the start. Neither court addressed what happens if a party violates AI restrictions - whether sanctions, waiver, injunctions, or other remedies apply.
Traditional remedies like clawback may not work once AI training has occurred. Prevention at the outset is the only reliable strategy.
Privilege Questions Remain Unsettled
The fundamental legal question - whether an AI tool is a "third party" that waives privilege or merely a "tool" that doesn't - remains unresolved. Both courts sidestepped this by using protective orders instead of privilege rulings.
Courts are taking a pragmatic approach: regulate AI through protective orders emphasizing control, prevention, and accountability while the privilege doctrine catches up. Until clearer rules emerge, assume AI use in discovery will be scrutinized.
Learn more about AI for Legal professionals and how to manage AI compliance in your practice. Paralegals handling discovery should review the AI Learning Path for Paralegals to understand these emerging restrictions.
Your membership also unlocks:
