Artificial intelligence applications such as ChatGPT, Google Gemini, Microsoft Copilot, and others are everywhere. Their use has grown at a remarkable pace over the past several years. What you may not realize, however, is that the information you share with these AI tools is not private. Your information, including queries and all responses, is stored by the companies that operate the apps and can be obtained by third parties, including the opposing side in your personal injury or civil rights lawsuit. Just as we advise clients to be careful about what they post on social media, we must now warn you with equal urgency: be extremely careful about what you say to an AI chatbot or app while your case is pending.
Why the urgency? Courts are beginning to address this issue directly. In a landmark ruling in United States v. Heppner, Judge Rakoff of the United States District Court for the Southern District of New York held that a party’s conversations with an AI assistant were neither protected by the attorney-client privilege, nor by the attorney work product doctrine. In that case, the defendant had used an AI application, without direction from his attorney, to research his case, draft strategy notes, and think through his legal situation. He later forwarded those AI-generated materials to his lawyer, believing they would then be shielded from disclosure as privileged. The court disagreed. Because the defendant had shared his confidential information with a third-party technology company (the AI app), which is a non-lawyer, and because that company’s own privacy policy expressly stated that user conversations could be stored, used to train the AI model, and disclosed to third parties in connection with claims, disputes, or litigation, the court found he had no reasonable expectation of confidentiality. Accordingly, those documents were subject to discovery by the opposing side in his case.
This ruling reflects a broader principle that attorneys and courts have long recognized: the attorney-client privilege is destroyed when a client shares otherwise confidential information with a third party. This point cannot be stressed enough – an AI chatbot or app is not your attorney. It is a product run by a private company whose terms of service authorize the collection and potential disclosure of everything you type. The fact that a conversation feels private does not make it legally protected. As the Heppner court confirmed, forwarding an AI-generated document to your lawyer does not retroactively protect it if it was not privileged in the first place.
Even if you have not asked the AI app anything directly about your case, the content of your prompts matters enormously. An experienced defense attorney can subpoena your AI chat history and use it to challenge the credibility of your injury claims, contradict your sworn testimony, suggest that your injuries have improved, question your motivation for bringing the lawsuit, or imply that you are coaching witnesses. Legitimate questions such as asking whether you can return to physical activity, how much a certain type of injury might be worth, or what happens if your case does not succeed, can be taken out of context and used against you in ways you would never anticipate. In addition, AI responses are very often wildly inaccurate and include “hallucinations.”
This means any advice that your lawyer gives you, which you put in the chat, is now discoverable. This behavior waives the protection of attorney client privilege and can be a real problem in your case. Moreover, AI is not a reasoning lawyer – it is a large language model, which does not dispense accurate legal advice. It can be useful for very general inquiries, but specific case and jurisdiction dependent strategy and value determinations from AI apps are not usually correct.
In Virginia, the discovery process is governed by the Virginia Supreme Court Rules, which give the party on the other side of your case quite a few tools to obtain your information. Pursuant to Virginia Supreme Court Rule 4:1(b)(1), any information sought in discovery only needs to be relevant or “reasonably calculated to lead to the discovery of admissible evidence.” Under Rules 4:9 and 4:9A, the opposing party can ask the court to force you to produce electronically stored information, which includes AI chat logs stored on a company’s servers. Because AI platforms like OpenAI (which operates ChatGPT), Google, and Microsoft are third-party companies that store your data, they can be subpoenaed to produce your conversation history, just as a social media company can be subpoenaed for your posts and messages.
There is an additional danger in attempting to address this problem by deleting your AI chat history. Most AI platforms retain your data on their servers for a period of time after deletion from your visible account. More importantly, once you are involved in a lawsuit or you reasonably anticipate litigation, you are under a legal duty not to destroy any potentially relevant evidence. Intentionally deleting AI conversations after that point could be treated as what is referred to as “spoliation of evidence.” This is considered bad conduct which can be used against you. If your case goes to trial, a spoliation instruction may allow the jury to assume that the destroyed evidence contained something damaging to your case. As we have advised with social media posts: do not delete anything. If you believe you may have already entered something harmful into an AI application, contact your attorney immediately.
While there are some uses of AI tools that carry little or no legal risk, such as searching for general health information unrelated to your specific injury, or using AI for tasks completely unrelated to your lawsuit, you should treat every AI application as a forum where the opposing attorney is reading over your shoulder. Never enter into any AI chatbot or app the following:
- a description of how your accident or incident occurred;
- information about your symptoms, pain levels, or recovery;
- anything your attorney has told you;
- questions about the value of your case or what you hope to receive;
- the names of any parties or witnesses involved; or
- any doubt, frustration, or concern you have about your case.
The law in this area is developing rapidly, and courts across the country are reaching similar conclusions. What is clear today is that AI chatbots are not lawyers, do not offer the confidentiality that many users assume, and that the legal system is fully prepared to treat your AI conversations as discoverable evidence. The same care and discipline that we urge all clients to apply to social media (do not post anything about your case) applies with equal force to everything you type into an AI application.
If you are injured through the fault of someone else, contact an experienced personal injury attorney at the Halperin Law Center, as we are familiar with both state and federal law. And remember: be careful what you say to AI.
