Inventors increasingly use generative AI tools to draft, refine, and organize their ideas before ever consulting counsel. A recent federal ruling on privilege offers a cautionary framework for evaluating the risks of that practice, particularly where consumer AI platforms use client inputs as training data.
The Heppner Ruling
On February 17, 2026, Judge Jed Rakoff of the Southern District of New York issued a memorandum in United States v. Heppner, holding that documents a criminal defendant created using Anthropic’s Claude were protected by neither the attorney-client privilege nor the work product doctrine. Heppner, after receiving a grand jury subpoena, used Claude on his own initiative to draft reports outlining defense strategy, then shared those reports with counsel and asserted privilege over them.
The court applied the traditional three-part privilege test: a communication, between privileged parties, made for the purpose of obtaining legal advice. The exchanges failed on multiple grounds. Claude is not an attorney, and the court noted that “in the absence of an attorney-client relationship, the discussion of legal issues between two non-attorneys is not protected by attorney-client privilege.” Because the communications were never privileged when made, sharing them with counsel afterward did not retroactively create privilege.
The confidentiality analysis is the portion most relevant to inventors. The court relied on Anthropic’s privacy policy, which states that Anthropic collects data on both users’ inputs and Claude’s outputs, uses that data to train the model, and reserves the right to disclose it to third parties, including government regulators. On that basis, the court found no reasonable expectation of confidentiality attached to the communications.
Application Beyond Privilege Disputes
The court’s reasoning is not confined to criminal proceedings. It rests on a general principle: information shared with a platform that reserves the right to retain, learn from, and disclose that information has not been kept confidential. That principle applies with equal force to inventors describing unfiled inventions to consumer AI tools.
A disclosure made without a duty of confidentiality can affect the timing and scope of available patent protection, including the novelty analysis for an inventor’s own later filing. A platform that retains and may disclose inputs presents a materially different risk profile than a closed conversation with counsel.
The court’s rejection of the argument that AI chat functions like ordinary software is instructive here. If a Claude conversation was not confidential enough to protect a legal strategy, similar terms would not render a conversation confidential enough to protect an invention disclosure.
Consumer Tools Versus Enterprise Deployments
Subsequent commentary on the ruling has drawn a distinction between consumer and enterprise AI products. One analysis observed that the decision “does not declare Gen AI incompatible with legal privileges,” and instead “applies settled doctrine to a new technology,” turning on the absence of confidentiality protections and attorney direction in that specific case.
That distinction is borne out in how the platforms are structured. Consumer tiers generally default to opt-out training consent, meaning inputs are used to train the model unless a user affirmatively disables that setting. Enterprise agreements, by contrast, typically exclude training use by default and impose contractual retention limits, with some platforms offering Zero Data Retention arrangements subject to account-level approval.
Recommendations
Counsel advising inventors on AI use should consider the following:
- Restrict consumer AI tools for unfiled inventions. Platforms whose terms permit training on inputs and disclosure to third parties present a genuine risk that information shared is not confidential, consistent with the court’s analysis in Heppner.
- Favor enterprise or contractually restricted tools. No-training provisions and defined retention limits provide stronger confidentiality assurances, though no arrangement guarantees absolute protection.
- Review data handling terms before disclosure. The provision that determined the outcome in Heppner was a privacy policy permitting training use and third-party disclosure. The same provision, in a different context, can complicate a patent filing.
- Involve counsel before substantive AI use. Where an invention disclosure is developed with AI assistance, documentation showing that the tool’s data terms were reviewed, and that use occurred under appropriate confidentiality safeguards, strengthens the inventor’s position.
Conclusion
The Heppner ruling addresses privilege, but its underlying logic extends to any confidential information shared with a platform that reserves rights to retain and disclose user inputs. Inventors should treat consumer AI tools accordingly, and counsel should confirm that any AI-assisted disclosure occurs on terms consistent with maintaining patent rights.
Justin Miller is a solo patent attorney. In 2025 he started his own law firm, Distinct Patent Law, after nearly 15 years of practice. His firm is located in Saint Petersburg Florida. He serves clients in Tampa Bay, and because patent law is federal, can serve clients all over the United States.