A cautionary US judgement, could it happen over here? And another with caveats that made it permissable.
Published March 30th 2026
Everything You Type Into AI Can and Will Be Used Against You (well almost). A Federal Judge Just Made That Official.
Published Mar 30, 2026
A criminal defendant typed his defense strategy into Claude. He shared details and asked the AI to help him build his case. And a federal judge just ruled every single one of those conversations is fair game for prosecutors to read, use, and weaponize against him in court.
This is the first ruling of its kind in the country. And whether you are a lawyer, a business owner, a CEO, or someone who has ever typed anything sensitive into an AI chatbot, you need to understand what happened here, why the judge ruled the way he did, and what you need to do right now to protect yourself and the people you represent.
The Case That Changes Everything
The case is United States v. Heppner, and Judge Jed S. Rakoff of the Southern District of New York handed down the ruling from the bench on February 10, 2026, followed by a written opinion on February 17. Bradley Heppner, the former chairman of GWG Holdings and CEO of Beneficient, stands accused of securities fraud, wire fraud, conspiracy, making false statements to auditors, and falsifying corporate records. Federal prosecutors allege he secretly controlled a shell company called Highland Consolidated Limited Partnership, fabricated debt obligations, and tricked GWG's board into funneling more than $150 million through that shell company for his personal benefit, using the money to fund his lifestyle, renovate his Dallas mansion, and improve his East Texas ranch. GWG eventually filed for bankruptcy, leaving tens of thousands of retail bondholders with losses exceeding $1 billion.
After Heppner received a grand jury subpoena and knew he was a target, he turned to Claude, the consumer version of the AI platform built by Anthropic. He took information his lawyers had shared with him and typed those details into Claude. He asked the AI to analyze his legal exposure. He generated reports outlining a defense strategy based on the charges his lawyers anticipated. He created 31 separate documents through his conversations with the AI. He then shared those AI generated reports with his defense counsel, and those reports shaped his lawyers' strategy going forward.
When the FBI executed a search warrant at Heppner's home upon his arrest, agents seized his electronic devices. Those devices contained all 31 AI generated documents. Heppner's lawyers put those documents on a privilege log, describing them as "artificial intelligence generated analysis conveying facts to counsel for the purpose of obtaining legal advice."
The government disagreed. Prosecutors filed a motion arguing the documents were not privileged and not protected work product. Judge Rakoff agreed with the government on every point.
Why the Judge Said No to Attorney Client Privilege
Judge Rakoff identified three independent reasons why attorney client privilege did not protect Heppner's AI conversations. Each one stands on its own. Together, they demolish any claim of privilege over consumer AI interactions.
First, Claude is not an attorney. Attorney client privilege requires a communication between a client and a licensed attorney. Judge Rakoff was direct about this. He wrote in his opinion that because Claude is not an attorney, that alone disposes of the privilege claim. No attorney client relationship exists between a user and an AI platform. Period. Full stop. The entire privilege framework requires a licensed lawyer on the other end of the conversation, and no AI chatbot meets that standard.
Second, there was no reasonable expectation of confidentiality. This is where the ruling gets personal for every single person who has ever typed something sensitive into a consumer AI tool. Judge Rakoff pointed to Claude's privacy policy, which expressly reserves Anthropic's right to disclose user data to third parties, including governmental regulatory authorities. The privacy policy also states Anthropic collects data on user inputs and the AI's outputs to train its models. Users must consent to this policy to use Claude. The judge reasoned this policy clearly puts users on notice their conversations are not private. You agreed to the terms. You handed your information to a third party. You destroyed your own confidentiality the moment you hit enter.
Third, the conversations were not made for the purpose of obtaining legal advice. Heppner's lawyers argued he used Claude for the "express purpose of talking to counsel." The judge rejected this argument. Heppner did not use Claude at the direction of his attorneys. His lawyers did not tell him to run those searches. He did this on his own. And when the government asked Claude whether the AI provides legal advice, Claude responded clearly that it is not a lawyer, cannot provide formal legal advice or recommendations, and recommended the user consult with a qualified attorney. The AI itself tells you not to rely on its output as legal counsel. The judge found the predominant purpose of Heppner's use of Claude could not possibly have been to obtain legal advice when the tool itself expressly warns against treating its output that way.
Judge Rakoff also made clear a fundamental principle every lawyer already knows. Sending unprivileged documents to your attorney does not make them privileged after the fact. Heppner shared his notes with a third party, Claude, before those notes ever reached his attorneys. Non privileged material does not become privileged because you later forward the material to your lawyer.
Why the Work Product Doctrine Failed Too
The work product doctrine protects materials prepared by or at the direction of counsel in anticipation of litigation. Heppner's lawyers never directed him to use Claude. He acted on his own initiative. The 31 documents did not reflect his attorneys' mental impressions or legal strategy. His lawyers were not involved in creating them.
Judge Rakoff was clear. Because Heppner created the documents without counsel's direction and the documents did not reflect defense counsel's strategy, the work product doctrine simply did not apply. You cannot create documents on your own, hand them to your lawyer later, and then claim your lawyer prepared them in anticipation of litigation. The doctrine requires attorney involvement from the beginning.
The Kovel Carve Out Lawyers Need to Understand
There is one narrow opening Judge Rakoff left in this ruling. He noted that if counsel had actually directed Heppner to use Claude, the analysis might have been different. Under the Kovel doctrine, a highly trained professional acting as an attorney's agent falls within the protection of attorney client privilege. The judge acknowledged Claude might arguably have functioned in that capacity if the lawyer had been the one directing the client to use the AI tool.
This distinction matters enormously. Attorney direction is the dividing line between protected and unprotected AI use. When a lawyer directs a client to use an AI tool as part of the legal representation, that use looks like a professional agent assisting counsel. When a client uses AI on their own, outside counsel's direction, the privilege framework collapses.
The Same Day Split That Makes This Even More Complicated
On the exact same day Judge Rakoff issued his bench ruling in Heppner, February 10, 2026, Magistrate Judge Anthony P. Patti in the Eastern District of Michigan reached the opposite conclusion in a civil case called Warner v. Gilbarco, Inc. Plaintiff Sohyon Warner, representing herself pro se, brought employment discrimination claims against Gilbarco, Inc. and Vontier Corporation. During discovery, the defendants moved to compel production of all documents and information related to Warner's use of third party AI tools in connection with her lawsuit.
Magistrate Judge Patti denied the motion. He ruled the AI generated materials were protected under the work product doctrine. His reasoning went in a completely different direction. He described ChatGPT and other generative AI programs as "tools, not persons, even if they may have administrators somewhere in the background." He stated no cited case orders the production of a litigant's internal mental impressions reformatted through software. He held the plaintiff's disclosure of information to ChatGPT did not amount to disclosure to an adversary or in a way likely to reach an adversary's hands, which is the standard required for work product waiver. He warned the defendants that their theory, if accepted, would nullify work product protection in nearly every modern drafting environment, a result no court has endorsed.
Two federal courts. The same day. Two different answers to the same fundamental question. Is an AI platform a third party or a tool?
Judge Rakoff treated Claude as a third party, like a friend or advisor you confide in outside the privilege relationship. Magistrate Judge Patti treated ChatGPT as a tool, like a word processor or research database. There is also a critical factual distinction between these cases. Heppner was a client who had retained attorneys and then independently used AI without their direction. Warner was a pro se litigant acting as her own counsel, making her use of AI inseparable from her litigation strategy and preparation. That distinction mattered in the analysis, and future courts are going to parse these factual differences carefully. This split is going to generate litigation across the country for years to come. And until appellate courts weigh in, every lawyer and every client is operating in uncertain territory.
The Enterprise AI Question Still Open
Judge Rakoff's opinion left one significant question unanswered. His analysis focused on the consumer version of Claude, where Anthropic's privacy policy permits data collection and potential disclosure to third parties. The judge expressly noted enterprise AI tools typically do not train on user data and may present a different analysis. Enterprise versions of AI platforms offer contractual confidentiality guarantees and assurances user inputs will not be used for training.
This distinction matters for law firms, corporate legal departments, and any organization using AI in connection with legal work. If you are running your sensitive legal analysis through a consumer AI chatbot, you are broadcasting your strategy to a third party with no confidentiality obligation. If you are using an enterprise platform with contractual protections, you at least have an argument your communications remain confidential. The judge did not decide that second question. Someone will litigate this soon.
The Smart Lawyer's Opportunity
Every lawyer reading this should see more than a cautionary tale. This ruling is an opportunity. An opportunity to get ahead of a problem before your clients walk straight into the same trap Bradley Heppner fell into.
Whether you like it or not, your clients are using AI right now. They are typing case details into ChatGPT, Claude, Gemini, and every other platform available to them. They are doing this at home, at work, on their phones, in bed at night, and they are doing all of this without understanding the legal consequences. They do not know their conversations are stored. They do not know the platform's privacy policy gives the company the right to review and share their data. They do not know a federal judge has now ruled those conversations are wide open for discovery.
You need to tell them.
Start with every new client you bring on. Make AI usage a standard part of your intake conversation, the same way you discuss document preservation, social media, and communication protocols. Ask them directly whether they have used any AI tools to research, analyze, or prepare anything related to their legal matter. Find out what they typed, what platform they used, and whether they used a consumer or enterprise version.
Go further. Reach out to existing clients right now and have this conversation. Send a letter. Pick up the phone. Get the message across clearly and specifically. Tell them what happened in Heppner. Tell them the risks. Tell them to stop inputting any case related information into consumer AI platforms immediately.
Update Your Retainer Agreements and Engagement Letters
Here is where proactive lawyers separate themselves from everyone else. Your retainer agreements and letters of representation need new language addressing AI use. Specifically, you need provisions covering the following.
A clear statement informing clients their conversations with consumer AI platforms like ChatGPT, Claude, and Gemini are not protected by attorney client privilege and are not covered by the work product doctrine if they use these tools without attorney direction.
An explicit instruction directing clients not to input any information related to their legal matter into any consumer AI platform without first receiving written approval from counsel.
A warning explaining consumer AI platforms collect and store user inputs and outputs, and these platforms reserve the right to share this data with third parties, including government authorities.
A statement making clear sending AI generated documents to counsel after the fact does not make those documents privileged.
Language addressing the distinction between consumer AI platforms and enterprise AI platforms, so clients understand the difference and know which tools carry greater risk.
A clause requiring clients to immediately disclose to counsel any AI generated materials they have already created in connection with their legal matter, so counsel can assess exposure and develop a strategy to address any privilege issues.
These provisions protect your clients. They protect your firm. And they create a documented record showing you addressed this issue head on.
Educate Everyone in Your Firm
This is not a partners only conversation. Every single person in your law firm needs to understand this ruling and its implications. Associates, paralegals, legal assistants, office managers, and support staff all need to know the rules.
If anyone in your office is using consumer AI to draft documents, research legal issues, analyze case facts, or prepare client communications, they need to understand those activities may destroy privilege and create discoverable material. Your firm needs an AI usage policy, and the policy needs to be specific about which platforms are approved for use, what types of information are permitted as inputs, and what safeguards must be in place before anyone touches an AI tool in connection with client work.
Run a training session. Make attendance mandatory. Walk through the Heppner ruling. Walk through the Warner ruling. Explain the split. Explain the risks. Give your team concrete guidance they follow starting today, not next quarter, not when it becomes more convenient.
Your Clients Are Counting on You
Every person facing litigation, every business owner in a dispute, every executive under investigation is looking for an edge. AI feels like an edge. You type in your problem, and the machine gives you analysis, strategy, arguments, and research in seconds. The temptation to use these tools is enormous, and your clients are already giving in to the temptation every single day.
Your job is to make sure they understand the cost. A federal judge in the Southern District of New York has now told the entire country these conversations are not protected. The privilege you and your client depend on to build a defense, negotiate a settlement, and prepare for trial evaporates the moment your client types case details into a consumer AI chatbot.
Share this ruling with your clients, your colleagues, and your professional network. Update your engagement letters today. Build AI usage protocols into your practice. Start the conversation right now, because the next Heppner case is already in the making, and the lawyer who prepares for this wins.
The lawyers who ignore this ruling and keep operating as if AI conversations are private are going to learn the hard way. And their clients are going to pay the price.
https://www.linkedin.com/pulse/everythi ... kson-bb3xf
