AI Isn’t Your Lawyer and That May Matter More Than You Think

Artificial intelligence tools are a part of everyday business life. AI can draft emails, summarize documents, organize information, and even help users think through potential legal arguments or strategies. And because AI can do those things quickly and efficiently, it’s not surprising that someone facing potential legal exposure might consult it before contacting counsel.

But a recent federal decision suggests that what AI can do and what it should be used for are not always the same – particularly when privilege and confidentiality are at stake.

In United States v. Heppner, Judge Jed Rakoff considered whether a defendant’s written exchanges with a publicly available AI platform were protected by attorney-client privilege or the work product doctrine. Applying traditional privilege principles – not creating new AI-specific rules – the court concluded that, at least in this instance and likely many others, they were not.

The opinion does not announce a categorical rule that anything typed into AI is discoverable. Instead, it walks through traditional requirements and asks a straightforward question: do communications with a third-party AI platform satisfy the elements of privilege?

In this case, the answer was unequivocally no.

When the Investigation and AI Use Collided

In October 2025, a New York grand jury charged Bradley Heppner, a financial executive, with multiple counts of fraud, conspiracy, and financial misconduct. Before the indictment was returned, Heppner had received a grand jury subpoena and knew he was a target of an ongoing investigation.

While the investigation was underway, Heppner began using the popular AI program, Claude. According to his counsel, he used the tool to generate possible defense arguments and to anticipate how the government might frame the case.

Importantly, his attorneys acknowledged that they did not tell him to use the AI platform. The decision to engage with Claude was entirely his own.

When federal agents searched Heppner’s home, they seized his electronics and other documents. Among the materials recovered were approximately thirty documents reflecting his written exchanges with Claude. Heppner tried to claim that these exchanges were privileged.

Heppner argued that the documents should be protected because he created them while anticipating he was going to be charged, incorporated information he had learned from his counsel, and later shared them with his attorneys as part of preparing his defense.

The government and the court disagreed.

Why the Court Rejected Heppner’s Privilege Claim 

Attorney-client privilege protects communications that are:

• With a licensed attorney (or their authorized agent) acting in an official capacity.
• Intended to be—and remain—confidential.
• Primarily for the purposes of obtaining legal advice (not business or personal).

It is the job of the party claiming privilege to show that all three of those elements are satisfied.

Unfortunately for Heppner, his defense fell short on every count, and the court ultimately ruled that his AI-generated legal advice did not qualify for the attorney-client privilege. Here’s a closer look at why.

Your Friendly Neighborhood Chat Bot Is Not an Attorney

The first element of privilege requires communication to have taken place between attorney and client. Heppner argued that since he shared the AI documents with his attorney, they automatically qualified as privileged. However, this is false.

Outside of limited exceptions, the attorney-client privilege does not extend to non-attorney communications. Communications with non-attorneys – including publicly available AI platforms – do not fall within privilege. And the fact that Heppner shared this information with his attorney post-facto doesn’t suddenly turn his computer into a human, let alone an attorney.

AI platforms are not licensed professionals. They owe no fiduciary duties, are not subject to attorney regulation, and expressly disclaim providing legal advice. The information they generate is not independently verified or even checked for mistakes, so whatever advice AI gives shouldn’t be taken at face value to begin with.

What You Type Into AI Doesn’t Stay in AI

Another problem with Heppner’s argument concerns privacy. Remember, only private communications that actually stay private qualify for privilege.

Heppner typed his queries into Claude, an AI tool created by an outside third-party. Claude drew its answers from multiple unverified sources, and, most importantly, (like virtually every AI platform) explicitly warned Heppner that the information it collects may be retained, used for training, and potentially disclosed to third parties, including the government.

In light of all of this, the court found that Heppner did not have a reasonable expectation that his communications would remain private.

No Such Thing as “Retroactive” Privilege

Even though Heppner later shared the AI-generated materials with his attorneys, the court emphasized a basic but important rule: sharing non-privileged material with counsel does not retroactively make it privileged. If a communication was not protected when it was created, it does not “become” privileged simply because someone eventually sends it to their attorney.

Notably, the court did not hold that all AI use destroys privilege. Instead, it applied the existing doctrine and concluded that the required elements could not be met. And given how common all of the issues that ruined Heppner’s argument are, it is difficult to imagine how a user could establish privilege without strict confidentiality controls and clear attorney involvement from the outset.

Why the Work Product Doctrine Also Failed

Heppner also invoked the work product doctrine. The work product doctrine protects materials prepared by or at the direction of counsel in anticipation of litigation – particularly those reflecting an attorney’s mental impressions, legal theories, or strategy.

Even assuming the AI documents were created while litigation was anticipated, the court concluded they did not qualify. Heppner’s lawyers admitted that they did not tell him to use Claude. Heppner consulted with Claude independently, not at his counsel’s direction, and the documents that were created did not reveal counsel’s strategy at the time they were created – even though Heppner did try to add in his attorneys’ advice.

The Court declined to expand the work product doctrine beyond its traditional purpose: protecting an attorney’s strategic thought process – not materials a client prompts a chatbot to create.

What This Decision Signals 

Judge Rakoff’s opinion is careful, but its direction is clear. Courts are unlikely to carve out new privilege protections simply because communication involves emerging technology. Instead, courts will apply the existing doctrine – strictly – and ask whether the required elements are actually met.

In Heppner, they were not.

That matters. The court did not need sweeping language to reach its conclusion. It simply applied settled principles: privilege requires a protected legal relationship and a reasonable expectation of confidentiality. Work product requires involvement by counsel.

A publicly available AI platform fits neither category.

As AI tools become more integrated into business and personal decision-making, this issue will arise again. And unless the factual circumstances are materially different – for example, where AI is used under counsel’s direction in a controlled and confidential setting – courts are likely to reach similar conclusions.

A Practical Reality

This decision is not an indictment of AI. It is a reminder about process.

If you are facing an investigation, regulatory inquiry, or potential litigation, using AI to test defenses, analyze exposure, or draft strategic outlines before speaking with counsel may feel efficient and cost-effective. But it introduces a third party into the communication – and that changes the privilege analysis.

Just because AI can generate legal-sounding analysis does not mean it should be used in place of, or prior to, a protected conversation with an actual attorney.

When legal exposure is involved, involving counsel first is not simply best practice – it is what preserves privilege in the first place.

AI is a powerful tool. It is not a protected legal advisor. As courts continue to address these issues, that distinction will likely remain decisive.