05/07/2026
Are we holding AI to a standard we don't apply to any other technology in our practice?
I've been having this conversation with several colleagues lately. The question is whether using an AI tool like Claude on client-adjacent work is an ethical risk. The biggest concern: the privacy carve-outs in most LLM's terms of service.
My response: those same carve-outs exist in Clio, Microsoft 365, and other cloud service providers. We've accepted analogous risk for years and never called it an ethics problem.
ABA Formal Opinion 512 (July 2024) gives us the framework. On confidentiality, it frames our obligation as a risk assessment rather than a guarantee of zero exposure. We evaluate:
— Likelihood of disclosure
— Sensitivity of the information
— Difficulty of implementing safeguards
— Extent to which safeguards interfere with the representation
That's a reasonableness standard. It's fact-specific. And reasonable lawyers can land in different places.
Here's what my risk assessment and implementation looks like in practice:
✅ Written firm policy on how client data is used with AI
✅ Paid subscriptions that opt out of training on day one
✅ AI-specific disclosures in my fee agreement
I'm not comfortable letting perfect be the enemy of the good. A zero-retention enterprise contract requiring 20+ seats is not a realistic option for a small firm. But that doesn't mean we can't use these tools responsibly and ethically.
Want to learn about how I use AI in my practice? This month, I am teaching 2 programs for the National Association of Consumer Bankruptcy Attorneys to show you exactly how to do it! May 14 & May 28 - $199 each for members | $399 non-members. Link in the Comments.
