An Important Addendum to “Can AI Draft Your Contracts: What the Benchmarking Really Says”
Last week, I wrote about research showing that AI contract drafting tools achieved 73% reliability compared to 57% for human lawyers on first drafts. The takeaway was clear: for routine transactional work, AI can be a powerful tool to accelerate the drafting process, provided human lawyers review and refine the output.
But a federal court ruling this week highlights a critical distinction that every founder and executive needs to understand: just because AI can help with some legal tasks doesn’t mean it should be your first call any time a legal issues arises—especially issues that might end up in court.
The Heppner Ruling: A Cautionary Tale
On February 10, 2026, Judge Jed Rakoff of the Southern District of New York ruled that 31 documents a defendant created using an AI tool (Anthropic’s Claude) and later sent to his lawyers were not protected by attorney-client privilege or work product doctrine. The defendant, Bradley Heppner, faced fraud charges and had used the AI to “prepare reports outlining his defense strategy and potential legal arguments” after receiving a grand jury subpoena and engaging counsel.
The government moved to compel production. The court agreed, holding the documents weren’t privileged because they failed every element required for protection.
Why the Documents Weren’t Privileged
The court’s reasoning rested on fundamental privilege principles that apply regardless of which AI tool you use:
The AI isn’t your lawyer. Privilege protects communications between attorney and client. An AI tool has no law license, owes no fiduciary duty, cannot form an attorney-client relationship, and explicitly disclaims providing legal advice. Talking to an AI isn’t talking to your lawyer, no matter how sophisticated the tool.
Pre-existing documents don’t become privileged by forwarding them to counsel. This is bedrock privilege law. If you Google “fraud defense strategies,” print the results, and email them to your lawyer, those printouts don’t become privileged. The same principle applies to AI-generated documents. The court analogized AI use to conducting library research or Google searches—forwarding the results to your lawyer doesn’t transform them into protected communications.
The client acted independently, not at counsel’s direction. Work product doctrine protects materials prepared by or at the direction of counsel in anticipation of litigation. Here, the defendant’s lawyers conceded he created the AI documents “on his own initiative, not at counsel’s direction.” That admission was fatal to the work product claim.
The Privacy Policy Red Herring
It is important to note that whether the AI tool has an enterprise grade privacy policy is likely irrelevant. Because the court raised the issue without deciding it, some of the commentary on the case has focused on the AI tool’s privacy policy, which permitted data collection and disclosure to government authorities. Some have suggested that using “enterprise” AI tools with better confidentiality protections would solve the privilege problem.
This is wishful thinking. The privacy policy issue is a sideshow. Even with a bulletproof enterprise agreement featuring zero data retention, no training on inputs, and contractual confidentiality obligations, you’d still have a non-lawyer tool disclaiming legal advice, creating documents that pre-exist transmission to counsel, and operating without counsel’s direction. The fundamental doctrinal problems remain.
The Practical Guidance: Know When AI Isn’t the Answer
The Heppner ruling doesn’t mean AI is worthless for legal work. But it does clarify when you should not reach for AI first:
Regulatory investigations or inquiries. If you receive a subpoena, CID, or regulatory letter, your first call should be to a lawyer, not an AI. Any documents you create before engaging counsel—including AI-generated summaries, timelines, or analyses—are likely discoverable.
Potential litigation. If you’re facing a dispute that might end up in court, involve human counsel from the outset. Creating “case strategy” documents with AI before talking to a lawyer doesn’t protect them—it creates discoverable evidence of your thinking.
Internal investigations. If you’re investigating potential misconduct, fraud, or compliance violations, structure the investigation under attorney direction from day one. Employee statements to AI tools lack privilege; employee statements to counsel in a properly structured investigation may be protected.
Sensitive transactions with litigation risk. Some deals carry obvious downside litigation risk—acquiring a company with pending lawsuits, negotiating settlements, restructuring around disputes. Get lawyers involved early, and don’t use AI to “organize your thoughts” before engaging counsel.
When AI Still Makes Sense
None of this changes the contract drafting analysis from my previous post. For routine transactional work—vendor agreements, NDAs, employment offer letters, standard licensing terms—AI-assisted first drafts reviewed by human lawyers can save significant time and cost.
The difference is the nature of the work. Transactional documents are created to be shared with counterparties. There’s no privilege to lose because the documents were never privileged in the first place. And the typical workflow (AI first draft → lawyer review → negotiation) means a human lawyer validates the output before it matters.
But litigation and investigations work differently. The whole point of privilege is to create a protected space for candid attorney-client communication. You can’t retroactively create that protected space by forwarding your AI research to counsel after the fact.
The Human Lawyer Remains Essential
The Heppner ruling is a reminder that technology hasn’t replaced—and won’t replace—the fundamental attorney-client relationship. Privilege doctrine developed over centuries to protect a specific professional relationship between a client and a licensed attorney bound by fiduciary duties and confidentiality obligations so that the client can rely on that protected space for candid discussion about her case.
AI tools are powerful research assistants, drafting aids, and efficiency multipliers. But they can’t replicate the attorney-client relationship, and pretending they can creates serious legal risk.
For founders and executives navigating the complex regulatory, compliance, and litigation landscape, the lesson is straightforward: when legal trouble is on the horizon, your first conversation should be with a human lawyer, not an AI.
This post discusses a bench ruling from U.S. v. Heppner, No. 25-cr-00503 (S.D.N.Y. Feb. 10, 2026). No written opinion has been issued yet, but the court’s oral ruling and the government’s motion make the reasoning clear.