Speaking to attorneys as the founder of OpenIntake, a voice‑AI company for law firms, I see that lawyers are rightly focused on what AI gets wrong. But far fewer are thinking about what AI remembers. 

Famous examples of AI safety failures in legal practice are already well known. In addition to the ‘Fake Citations’ news headlines we all know, there are now over 600 documented instances of lawyers sanctioned for hallucinated content in the United States alone. But what has received far less attention is how self-learning AI tools can retain and later surface confidential client information.

In Formal Opinion 512, the American Bar Association outlines this risk:

Self-learning Generative AI (GAI) tools into which lawyers input information relating to representation, by their very nature, raise the risk that information relating to one client’s representation may be disclosed improperly, even if the tool is used exclusively by lawyers at the same firm. This can occur when information relating to one client’s representation is input into the tool, then later revealed in response to prompts by lawyers working on other matters, who then share that output with other clients, file it with the court, or otherwise disclose it. In other words, the self-learning GAI tool may disclose information relating to the representation to persons outside the firm who are using the same GAI tool.

Still, most lawyers are not aware of which AI products are “self-learning”. In this article, I will clarify what this means and how to protect your firm as you adopt AI tools in 2026.

I. What exactly does AI “remember”?

Consumer AI tools like ChatGPT, Google Gemini, and Claude can “remember” information in two fundamentally different ways, with very different implications for client confidentiality.

First, ChatGPT includes a “memory” feature that saves limited information such as preferences or helpful context to the user’s account. These memories are private to that account and are intended to improve continuity.

The second type of “remembering” is largely hidden and poses the most risk. In ChatGPT, what you type may be used to “train” future versions of the system. When that happens, confidential information is no longer confined to your account and could potentially resurface to other users, which is why it raises serious confidentiality concerns for lawyers.

In contrast to general consumer tools like ChatGPT, legal-specific AI tools typically do not have any type of self-learning functionality, specifically for the purpose of protecting confidential information.

II. How can I protect my firm?

Fortunately, protecting confidential client data is an easy step that already works with your existing tools and carries no additional cost.

If you use ChatGPT for your legal work, you can go to Settings > Data Controls and toggle off the setting titled "Improve the model for everyone". This will prevent OpenAI (the developer of ChatGPT) from using your data to “train” future versions of ChatGPT. Similar settings exist for other consumer apps like Google Gemini and Claude.

If you use AI tools designed specifically for legal practice such as LexisNexis, Clio, or OpenIntake, you are in the clear. Industry-specific tools are often protected by contractual restrictions on data use, but this should be confirmed by reviewing your provider’s privacy terms.

Finally, firms should actively monitor employee use of consumer AI tools to ensure these protections are enabled and consistently applied.

III. Moving forward with AI

As your law firm begins adopting AI tools, safety should be a key consideration in protecting both your firm and its data. The safest options will consistently be legal-specific tools designed with the unique confidentiality requirements of legal professionals in mind. As the founder of a legal AI company, I am always open to discussing the latest developments in legal AI and welcome outreach by email at grayson@openintake.ai or by phone at 210-284-7966.

Grayson Pike is the founder of OpenIntake, a voice AI company for law firms. OpenIntake provides 24/7 client intake using natural phone conversations and is currently serving members of NJAJ. Grayson can be reached by email at grayson@openintake.ai or phone at 210-284-7966.