In last month’s newsletter, we discussed a federal case from New York which held that an individual’s artificial intelligence-generated documents were not protected by the attorney-client privilege. This has caused most attorneys to advise their clients not to use “public” AI tools for confidential matters without first consulting an attorney. Attorneys themselves are now worried about the second question, what happens when an attorney uses such AI tools for privileged matters.
It remains somewhat of an open question whether attorney-client privilege extends to third-party agents whose assistance is necessary to the attorney’s rendering of legal advice. Therefore, even attorneys have to be careful about using any AI tool on client matters, and to review the platform’s actual data practices to determine whether prompts are retained, whether they can be accessed, how they provide a response to legal process, and whether deletion is available.
As mentioned in the last newsletter, the Florida Bar published an excellent document on March 16, 2026, entitled “The Florida Bar Guide to Getting Started with AI.” It discusses each type of “general” AI model, including Anthropic’s Claude, Google’s Gemini, Microsoft’s Copilot, Open AI’s ChatGPT, and xAI’s Grok, along with law-specific AI models such as CoCounsel by Westlaw, Lexis Plus AI, and Vincent by Clio. It suggests that a good start is to first explore general AI models such as ChatGPT, Gemini, or Copilot for administrative tasks and brainstorming. It suggests that the legal AI models such as CoCounsel, Lexis Plus AI, or Vincent by Clio are better for legal research and document review. Visitors should be careful initially and not use any confidential or client-specific information, until the user is sure of the various legal protections.
In general, free general AI models use your questions and uploaded documents to train future models. Therefore, to maintain client confidentiality, you will need a paid subscription. You should confirm that your account settings do not allow data sharing and that you can delete your prompts and uploads at any time. Some platforms have methods to turn off or opt out of data sharing. One advantage of the three main legal AI models is their promise to maintain client confidentiality, use of strong security, and reduced hallucination risk. Remember that only certain paid licenses for general AI models promise prompts and uploads will not be used for training or stored. Remember also that the models are being updated and revised so that the user must generally rely on the current representations of each provider. The Florida Guide does a good job of discussing the relative merits of each type of provider.
The legal issues are so complicated that at least five federal district courts have issued specific protective orders restricting the use of generative AI platforms with protected materials. A detailed framework was drafted by the U.S. District Court of Colorado in Morgan v. V2X Inc., 2026, WL 864223 (D. Colo. 5/30/26). This particular Order includes the following:
No party or authorized recipient may input, upload, or submit CONFIDENTIAL Information into any modern artificial intelligence platform, including any generative, analytical, or large language [model-based] tool (‘AI’), unless the AI provider is contractually prohibited from: (1) storing or using inputs to train or improve its model; and (2) disclosing inputs to any third party except where such disclosure is essential to facilitating delivery of the service. Where disclosure to a third party is essential to service delivery, any such third party shall be bound by obligations no less protective than those required by this Order. In addition, the AI provider must contractually afford the party or authorized recipient the ability to remove or delete all CONFIDENTIAL Information upon request. A party intending to use AI that it contends meets these requirements must retain written documentation of these contractual protections.
Editor’s Note: The advice in this Florida Bar Guide is primarily directed towards lawyers, but many of the suggestions also apply to employers’ own use of AI platforms. Employers have the same interest in protecting their confidential information or trade secrets, as well as hopes of avoiding disclosure to opposing parties in court discovery.
One of the reasons for all this interest is that pro se (individual parties to litigation without legal representation) plaintiffs are extensively using AI for self-represented filings. Therefore, attorneys defending employers in employment litigation should consider seeking an AI order in the litigation to reduce the time spent verifying whether the cites or authorities exist and whether they support what the plaintiff claims. Defendant employers are also increasingly incorporating AI usage questions into their standard discovery requests, to identify what tools were used and whether histories were deleted or disabled, and request prompts and outputs tied to the drafting of pleadings or motions and to the factual narratives on which the pro se party relies. It is unknown at this point whether AI-assisted litigation preparation materials will be considered by the courts as protected work product. Further, lawyers may need to explicitly include AI information in their document preservation letters to their clients, as well as to the plaintiff. In general, until this area of the law is clarified, such information should be treated as general electronic information. Employers must also be counseled about any uploading of confidential information into AI tools.
These issues also are a further reminder that employers need AI usage policies to specifically address which platforms are approved with company data, prohibit the connection of personal AI agents to internal systems, and make clear that approved tools mean enterprise-tier deployments for the appropriate data handling function. If the company allows employees to use different systems to connect personal AI agents to those systems, the other party will argue the company failed to protect the very information it now claims is secret.
This article is part of our June 2026 Newsletter.
View the newsletter online
Download the newsletter as a PDF
