Problem: Your Employees Are Typing Legal Questions Into Public Chatbots

The problem

In United States v. Heppner, Judge Jed Rakoff of the Southern District of New York ruled (orally on February 10, 2026, with a written opinion on February 17) that documents a defendant generated using a consumer AI chatbot were protected by neither attorney-client privilege nor the work product doctrine.

The reasoning is worth understanding, because it maps directly onto what your employees do every day:

• The AI isn't a lawyer. No attorney-client relationship, no privilege.
• The exchange wasn't confidential. The platform's consumer terms allowed the provider to retain and potentially disclose data.
• He did it on his own. The defendant used the tool on his own initiative, not at counsel's direction, which sank the work product argument.
• Sharing it with counsel later didn't help. A non-privileged document doesn't become privileged because you forward it to your lawyer.

Not every court will land in the same place. A federal court in Michigan protected a pro se litigant's AI-assisted materials as work product the same month. But Heppner is the ruling everyone is citing, and "maybe a different judge will see it differently" is not a risk strategy.

Every time someone asks a public chatbot "Can we fire this person?" they may be writing Exhibit A.

The diagnosis: where the exposure actually lives

Run this quick check on your organization:

• Shadow AI usage: Do you know which AI tools employees use? (Hint: it's more than the ones you bought.)
• Consumer vs. enterprise: Are your approved tools on enterprise terms that bar training on inputs and restrict retention and disclosure?
• Direction of counsel: When business teams use AI for legally sensitive analysis, is it at legal's direction and documented as such?
• Policy clarity: Does your AI policy explicitly tell employees not to put legal matters into unapproved tools?

If you answered "not sure" to two or more, you have the problem.

The solution, part 1: technology

Give people a better door than the public chatbot

People use consumer AI because it's there and it's easy. The fix is to make the approved option easier. An enterprise-grade generative AI legal assistant with no-training commitments, data retention controls, and audit logs gives you a defensible foundation for confidentiality.

Route legal questions to legal

A self-service legal intake front door, ideally one the business can reach from Slack, Teams, or email, captures sensitive questions inside a counsel-directed channel instead of a consumer app. Bonus: you finally see what the business is worried about.

Keep the record in a system you control

Store AI-assisted legal work in your document management or matter system with proper privilege labeling, not in someone's chat history.

The solution, part 2: policy

1. Name the approved tools. A short list beats a long list of prohibitions.
2. Say it plainly. "Do not enter legal questions, disputes, investigations, or privileged communications into unapproved AI tools."
3. Direct the work. When legal asks a business team to use AI for analysis in anticipation of litigation, document that direction.
4. Train with examples. Show real scenarios: the HR complaint, the vendor dispute, the regulator's letter.
5. Update your litigation hold. Consider whether AI chat histories and prompts are in scope for preservation. Your legal hold tool should be able to reach them.

The bottom line

Heppner didn't create new law; it applied old privilege principles to a new tool. That's exactly why it matters. The fix is equal parts technology (an approved, enterprise-grade path) and policy (clear rules and counsel direction). Compare enterprise AI assistants and intake tools in the CorporateLegal.tech directory and close the gap before opposing counsel finds it.

This article is general information, not legal advice.