Are Your AI Prompts Privileged?

All forms of privilege rely on confidentiality as the bedrock of their protection.  With AI tools becoming more prevalent in our daily lives, it asks the question, what kind of protections do your communications have when using an AI tool?   

Recently, Sam Altman from ChatGPT indicated that requests to a standard ChatGPT engine lack legal privilege, indicating that his tool is akin to an attorney, priest, or therapist, who would be able to withhold information from the legal system.   These searches and interactions could be subpoenaed and might convince judges that any communications or information entered into ChatGPT are no longer confidential, which would break attorney-client privilege.   

Confidentiality is the cornerstone of attorney-client privilege and the critical concern here.  To protect attorney-client privilege, attorneys have a duty to ensure that communications and information with clients are confidential, such as avoiding having unnecessary parties as part of conversations, ensuring no outside parties are included on E-Mails chains to keep the materials limited and confidential.  When using an AI model, the question becomes whether entering facts or other information is different than using a traditional system, like Lexis or Westlaw, which are commonly used for legal research.   

ChatGPT and other open AI systems use your inputs to both identify an answer and help the system learn to provide additional answers in the future.  As this information is both stored externally by a company outside of your control, unlike a vendor with specific contracts and relationships, and could resurface in information used by other attorneys or members of the public by asking questions to the same system, there is a legitimate concern that these communications are no longer confidential, and therefore not protected by attorney-client privilege.  So, while there are tools on the market which will increase your efficacy as advocates, it is critical to ensure you are taking the right steps to ensure confidentiality.   

But with the proliferation of these models and options, how do you know what to look for with regard to maintaining confidentiality.  The primary way to confirm confidentiality is kept is to ensure your data is only used to only answer your question and is then removed from the system.  This will help alleviate concerns that confidentiality is broken or your sensitive materials are being used to enrich others.   

AI models which meet these criteria are likely to be proprietary and was trained on data in advance.  Rather than a general AI solution, these will be more likely tailored to your needs, whether they are in drafting, discovery, or research.  These are likely be paid AI models within a closed system, due to their proprietary nature and the fact they are not able to use your data to build further models.  However, the contractual phase will allow you to confirm the use and retention of your data, the presence of a closed system, and confirm the data security of the system to ensure your client’s data remains confidential and no privilege is broken through the use of efficiency improving AI tools.