Artificial intelligence (AI) offers clear efficiencies, but government contractors should proceed with caution before inputting protected documents, confidential information, or dispute-related facts into consumer AI tools in the hopes that they will quickly analyze your dispute and provide a strategy for recovery or litigation. A prompt intended to save time may instead produce unreliable analysis, expose sensitive information, create compliance problems under the contractor’s legal and contractual obligations, or jeopardize privilege. This article highlights several reasons government contractors—especially those facing disputes or involved in REAs, claims, and appeals—should think carefully before using AI to analyze sensitive matters.
AI’s Accuracy Problem
The most immediate concern is also the simplest: AI outputs can be wrong. As AI use becomes more common, so do the risks of error. One of the most persistent problems is that AI tools can generate responses that sound authoritative but are simply wrong. In October 2025, Deloitte faced scrutiny for using AI to help prepare reports for the government that reportedly included fabricated references (so-called “hallucinations”) to support their conclusions.1 As our Bid Protest Group recently advised, the lesson is straightforward: AI can be useful, but it cannot be trusted without careful human verification.
When Private Facts Become Public
A study done by the Stanford Institute for Human-Centered AI in 2025 compared the AI privacy policies of six major AI developers2 and found that all six of them not only utilized the users’ chat data to train their programs, but some even kept this information indefinitely.3 This matters because information entered into an AI platform may be retained, used to improve the model, or otherwise handled in ways that are outside the user’s control, increasing the risk of exposure to third parties, including adversaries and competitors.
Disputes and litigation often revolve around facts a company would never choose to air publicly: internal emails, pricing judgments, compliance missteps, performance problems, and other sensitive details that can damage customer relationships, invite government scrutiny, and create lasting reputational harm. Once sensitive information is entered into an AI platform, you lose control over where that information goes, how it is stored, who can review it, and whether it could later be exposed in ways you never intended. A prompt intended to save time may instead become the mechanism by which your sensitive and confidential information is disclosed beyond your control.
The stakes are even higher for government contracts where disputes, REAs, claims, investigations, and litigation often involve exactly the kinds of facts a contractor cannot afford to expose. In addition to internal analyses, pricing and cost data, performance issues, compliance concerns, and communications with counsel, there may be facts tied to CUI, source selection information, or other government-protected material. Depending on the information disclosed and the applicable contract requirements, careless use of AI may create exposure under confidentiality restrictions, cybersecurity obligations, CUI handling rules, source selection protections, and other regulatory or contractual requirements. In the dispute context, that can compound litigation risk with compliance risk.
AI’s Threat to Privilege
Using AI to analyze dispute fact patterns or perform case analyses can also jeopardize privilege and work product protections in later litigation. Attorney-client privilege protects confidential communications between attorney and client made for the purpose of obtaining or providing legal advice. Confidentiality is the core requirement. Once those communications or the underlying facts are shared with an AI platform, the privilege analysis becomes far less certain. Whether protection is preserved may depend on the tool used, the platform’s data-handling practices, the jurisdiction, and the circumstances under which the information was entered.
The emerging case law is cautionary but not yet settled. In United States v. Heppner, Judge Rakoff held that a criminal defendant’s exchanges with a public AI platform were protected by neither the attorney-client privilege nor the work product doctrine. The court emphasized that the defendant used the tool on his own—not at counsel’s direction—and under terms that undermined any reasonable expectation of confidentiality. By contrast, in Warner v. Gilbarco, Inc., the Eastern District of Michigan held that a pro se litigant’s AI-assisted materials were protected work product, reasoning that AI is a tool, not a person, and that work product is not waived absent disclosure to an adversary or a substantial likelihood that it will reach one. The key takeaway is not that AI use is categorically permissible or categorically disqualifying. It is that the privilege and work product analysis will likely turn on the facts—who used the tool, for what purpose, under whose direction, and on what confidentiality terms.
Key Takeaways for Government Contractors
The privilege, privacy, disclosure, and accuracy risks associated with AI case analyses are substantial, particularly where a public-facing platform may retain prompts, use them for training, or reserve broad rights over user data. For government contractors, the prudent course is straightforward: do not input sensitive dispute facts, legal strategy, privileged communications, CUI, source selection information, or other protected material into consumer AI platforms. Whatever efficiencies AI may offer, they rarely justify the legal, contractual, and business risks that can follow from disclosure.
____________________
1Accounting Times, Partner responsible for Deloitte AI bungle reportedly leaving the firm, (Dec. 5, 2025), https://www.accountingtimes.com.au/profession/partner-responsible-for-deloitte-ai-bungle-reportedly-leaving-the-firm.
2Amazon (Nova), Anthropic (Claude), Google (Gemini), Meta (Meta AI), Microsoft (Copilot), and OpenAI (ChatGPT).
3Stanford University Human-Centered Artificial Intelligence, Be Careful What You Tell Your AI Chatbot, (Oct. 15, 2025), https://hai.stanford.edu/news/be-careful-what-you-tell-your-ai-chatbot.
4Special thanks to Christian Pickard, a student intern in PilieroMazza’s Government Contracts Group, who assisted with this blog.
See more »
DISCLAIMER: Because of the generality of this update, the information provided herein may not be applicable in all situations and should not be acted upon without specific legal advice based on particular situations. Attorney Advertising.
© PilieroMazza PLLC
Refine your interests »
Join more than 70,000 authors publishing their insights on JD Supra
Back to Top
Explore 2026 Readers’ Choice Awards
Copyright © JD Supra, LLC
