Bench & Bar September/October 2025

NST

ARE THE PROTECTIONS SUFFICIENT? Various LLMs have confidentiality provi sions, but are they enough? Do you clearly understand what they are? Can you trust providers not to violate those provisions, whether inadvertently or otherwise? That very thing happened in 2024 when Microsoft disclosed contract provisions allowing employees to retain and review certain prompts falling into categories like hate speech or violence. Google and Claude have revealed similar practices. How can you be certain that your communications with an LLM aren't being reviewed by humans? Enterprise systems designed for legal use appear to offer greater protections, but even these providers might be forced to pro duce inputs and outputs when faced with a subpoena. Remember: most LLMs retain material indefinitely. Inputs and outputs don't just disappear. WHAT LAWYERS SHOULD DO Courts and rule-making bodies will need to grapple with these issues soon, and the results may be inconsistent. In the mean time, we must educate our clients about these risks through training sessions, just as we did with email and other technologies when they emerged.

As attorneys, we need to be thoughtful about what we input into LLMs and how we use GenAI tools. Consider the impli cations if your inputs and outputs became discoverable. Think through privilege and waiver issues. Read the terms and condi tions carefully. I'm a strong advocate for using GenAI tools—they offer tremendous potential ben efits for the legal profession. But we cannot ignore the risks that come with their use. BEST PRACTICE: THE NEW YORK TIMES RULE The best advice comes from a trial lawyer and colleague: always follow the New York Times rule. Don't put anything in an LLM that you would be uncomfortable seeing in a New York Times article. Courts are beginning to address these issues, and the landscape is evolving rapidly. Until we have clearer guidance, prudence should guide our approach to these pow erful but potentially risky tools. A tip of the hat to Judge Scott Schlegel, whose post highlighted this issue and its potential dangers. His insights continue to provide valuable guidance for the legal profession as we navigate these emerging technologies.

ABOUT THE AUTHOR

STEVE EMBRY is a frequent speaker, blogger and writer. He is publisher of TechLaw

Crossroads, a blog devoted to legal technology and the practice of law. He also writes for Above the Law. He serves as co-chair of the Kentucky Bar Asso ciation’s Law Practice Committee and Artificial Intelligence Task Force. He is past chair of the ABA Law Practice Divi sion and co-chair of ABA TechShow. He was formerly a member of Frost Brown Todd LLC.

29 bench & bar