Generative AI Does Not Eliminate Discovery Burden: Key Lessons for White Collar Practitioners

Generative artificial intelligence AI (gen AI) promises greater efficiency in document review, but instead of eliminating discovery disputes, it has transformed them. Courts are now confronting questions about how much detail opposing counsel is entitled to about the use of gen AI tools. White collar practitioners face a parallel challenge: The government, increasingly sophisticated with new technology, is scrutinizing companies' use of gen AI and other AI tools in discovery. Prosecutors often demand details on technology-assisted review (TAR) protocols, ask probing questions about AI methodologies and sometimes insist that companies apply gen AI before narrowing the review population with search terms.

The fundamental tension is clear: Though gen AI is making discovery faster and sometimes cheaper, it has not made discovery free. Collecting and storing massive amounts of data remains costly and time-consuming and, for certain types of data and devices, intrusive to individuals' personal lives. Companies reasonably use AI to reduce discovery costs, not to expand its scope. Yet, opposing parties and the government sometimes treat AI tools as though they minimize burden concerns. A recent decision from the U.S. District Court for the Northern District of California pushes back on that assumption.

In Schulte v. LinkedIn Corp., No. 22-cv-00237-HSG (LB), 2026 WL 1905851 (N.D. Cal. June 30, 2026), the court delivered a resounding win for the defense on three key issues:

1. Gen AI is TAR. The court treated gen AI – specifically, the Relativity aiR tool – as a form of TAR governed by existing case law, rather than a novel technology requiring new procedural rules.
2. Search Terms First, Then Gen AI. The defendant was permitted to apply search terms before applying gen AI review. The court recognized that requiring gen AI review of all data without prior culling "would impose a substantial burden" given the volume at issue– approximately 800 GB for just two of 19 custodians – and acknowledged "significant costs related to processing" and "hosting."
3. Limited Disclosure of AI Methodology. The defendant was not required to provide detailed information about its use of gen AI, such as "elusion estimates, the document error rate, and the number of human reviewers being used to validate" the model's predictions. The court explained that "discovery on discovery" is disfavored and generally not relevant or proportional to the needs of the case. It was sufficient that the defendant disclosed: 1) there was no seed set or training used, 2) Relativity aiR was being used to make final responsiveness calls and 3) quality control review was being conducted by human review of samples taken from each responsiveness type.

Lessons for White Collar Practitioners

The Schulte decision empowers civil litigants to resist discovery overreach and provides a useful road map – and a helpful citation – for companies navigating government investigations. Though companies in the government investigation context have fewer procedural guardrails than civil litigants who can negotiate electronically stored information (ESI) protocols, the decision supports several practical approaches:

Push Back on Overbroad Demands. As discussed in a previous alert, the U.S. Department of Justice (DOJ) recently emphasized efficiency in white collar enforcement. In its 2025 policy memorandum, DOJ publicly acknowledged that "federal investigations into corporate wrongdoing can be costly and intrusive for businesses" and called for prosecutors to "maximize efficiency" to reduce that burden. Schulte reinforces that gen AI should create efficiencies, not expand discovery scope. When the government demands gen AI review of unculled data, counsel can cite Schulte, alongside the 2025 policy memorandum to push for a more proportional approach.

Negotiate Protocols Before Implementing Them. It is far cheaper to incur discovery costs once than to redo discovery steps because the government is dissatisfied with the original methodology. Counsel should negotiate gen AI protocols before enacting them wherever possible.

Thoroughly Document AI Protocols. Counsel should keep contemporaneous records of gen AI protocols – including the tool used, workflow decisions and quality control measures – so that if this information must eventually be produced, there is no need to recreate it after the fact.

For more information or questions, please contact the authors.

Information contained in this alert is for the general education and knowledge of our readers. It is not designed to be, and should not be used as, the sole source of information when analyzing and resolving a legal problem, and it should not be substituted for legal advice, which relies on a specific factual analysis. Moreover, the laws of each jurisdiction are different and are constantly changing. This information is not intended to create, and receipt of it does not constitute, an attorney-client relationship. If you have specific questions regarding a particular fact situation, we urge you to consult the authors of this publication, your Holland & Knight representative or other competent legal counsel.