Artificial intelligence is rapidly changing how lawyers analyze information. For federal investigations, that raises a question with two sides. Can the government adopt AI quickly enough to keep pace with sophisticated defense teams? And will the defense be allowed to use AI on the evidence the government produces?

The AI gap may run in either direction. The government may lag in adopting the technology. But it largely shapes the terms on which the defense can use discovery.

Imagine a corporate investigation involving millions of emails, texts, financial records, and contracts. AI tools can help lawyers sort that volume, identify relevant communications, build chronologies, and surface connections that would otherwise be difficult to spot.

AI can help find the needle. It can also help show what the haystack means. The question is which side will be equipped, and permitted, to use it.

The Government’s Side of the Gap

This is not a story about the government ignoring technology. DOJ's 2025 AI Use Case Inventory contains 315 entries, a 30.7% increase over the 2024 inventory. The Department says this reflects growing collaboration to "accelerate AI adoption and innovation."

In a September 19, 2026 media interview, FBI Director Kash Patel said the Bureau's use of AI to triage threats had increased by 605% over the previous 18 months. "AI, when used lawfully," he said, "is a critical tool to triage data."

Still, the government faces institutional challenges. A late-2024 DOJ Office of Inspector General audit found the FBI to be in the "early stages of AI integration." OIG credited the Bureau with initial steps, including designating a Chief AI Officer and publishing an AI policy directive. But it also identified barriers to adoption: limited funding, difficulty hiring and retaining technical staff, the need to modernize data architecture and IT infrastructure, AI embedded in commercial products that the FBI cannot verify without technical details vendors typically do not provide, and a shifting policy landscape that could create a backlog of AI use cases awaiting approval by the Bureau's AI Ethics Council.

The FBI's challenges are only part of the picture. Neither the Department's inventory nor the OIG audit tells us what is happening on the line, in the 94 U.S. Attorneys' Offices where most federal cases are actually investigated and tried. There are signs of movement: DOJ has deployed AI tools from LexisNexis, Westlaw and Bloomberg for legal research and has signaled interest in piloting an AI legal assistant. Research tools, however, are not the same as AI-assisted review and analysis of a case's discovery.

That raises a practical question: will the line AUSA handling a document-heavy fraud case have the same tools as the defense team across the table? A well-resourced defense firm can license an AI review platform for a single matter, and many firms have already deployed AI tools firmwide. A prosecutor may have to wait for Department-wide procurement, security approval, and training. The risk is not that prosecutors lack judgment. It is that they may be asked to exercise it over more data, with fewer tools, than their opponents. In a corporate case, where defense counsel collects and reviews the company's data, that could mean the defense understands the government's evidence better than the government does.

The Defense’s Side of the Gap

The picture changes when discovery enters it.

Courts are already deciding whether protective orders should restrict the use of discovery materials with AI, and they have taken different approaches. In Fiorito v. Metropolitan Council, 2025 WL 1806612 (D. Minn. July 1, 2025), the court entered a blanket prohibition: "No party may upload or otherwise disclose Confidential Data to any generative artificial intelligence platform, such as (but not limited to) ChatGPT."

Two federal criminal cases took a narrower approach. In United States v. Jackson, 2025 WL 1666249 (W.D. Wash. 2025), and United States v. Navarro Hernandez, 2025 WL 3290753 (W.D. Wash. 2025), the protective orders barred loading protected material into any AI tool where data rights are retained by a party that has not agreed to be bound by the order. That approach focuses on who controls the data, not on whether AI is used. It permits tools that contractually commit to the order's protections.

Members of the federal defense community have raised concerns about overbroad restrictions on defense use of AI. They urge that “defense counsel should vigorously resist these efforts using all available legal and ethical means.”

There are legitimate reasons for restrictions. Sensitive discovery should not lose its protections because someone uploaded it to a public AI platform. Courts have pointed to the practical impossibility of clawing back data once an open tool has taken it in.

But prohibiting disclosure to an unsecured AI system is different from prohibiting AI-assisted analysis altogether. A protective-order dispute may concern more than data security. It raises a more fundamental question: should one side be able to use AI to analyze the evidence when the other cannot?

AI Should Enhance, Not Replace, Judgment

None of this means AI should replace investigators or prosecutors.

AI might identify a connection, a communication or a pattern. Investigators still have to decide whether it means anything. Prosecutors still have to evaluate credibility, intent and context, and decide whether prosecution is warranted.

The goal is not AI instead of investigators. It is better-equipped investigators and lawyers on both sides of the process.

Practical Takeaways

• Raise AI terms early. Address AI use when the protective order is negotiated, not after discovery begins.
   
• Propose data-rights language rather than blanket bans. The Jackson and Navarro Hernandez formulation protects sensitive material without prohibiting secure AI-assisted review.
   
• Document your vendor's commitments. Know what confidentiality, retention and training terms your AI tools actually provide, and be prepared to show them to the court.

So, who will hold the advantage? That may depend less on who has the best AI than on who is permitted to use it, on what evidence, and under what safeguards. The answer should be both sides.