For the past two years, one of the biggest questions surrounding generative AI has been whether conversations with AI systems could enjoy the same legal protections as conversations with lawyers. A landmark US court decision has now delivered an answer.
In US v Heppner, a federal judge ruled that documents created by a defendant using Anthropic’s Claude AI were not protected by attorney-client privilege or the work product doctrine. The decision is the first of its kind and could have big implications for how law firms govern the use of AI, both internally and by their clients.
A case that could reshape legal AI?
The case centred on Bradley Heppner, a former CEO facing multiple fraud charges. Before his arrest, Heppner used Claude to analyse the government’s investigation, explore possible defence arguments and develop legal strategies. Believing these materials would be protected, he later shared them with his lawyers.
Instead, prosecutors successfully argued that the AI-generated documents should be disclosed. Even more damaging, Heppner’s own AI prompts were later used as evidence against him during trial, where he was convicted on all counts.
Why the court rejected privilege
The court’s reasoning was that Claude was not a lawyer, meaning no attorney-client relationship existed between the defendant and the AI system. Also, the conversations were not considered confidential because they had been shared with a public commercial platform whose privacy policy allowed prompts and outputs to be retained, used for training and, in certain circumstances, disclosed to third parties or government authorities.
Finally, the work was not created at the direction of legal counsel, meaning it did not qualify for protection under the work product doctrine.
But the most significant part of the judgment was not what the court rejected, but what it suggested.
Judge Jed Rakoff noted that if the AI had been used under the direction of counsel, the outcome might have been different. That single observation is likely to become one of the most closely analysed aspects of the decision because it hints that AI itself is not necessarily incompatible with privilege. Instead, the determining factor may be how AI is used, who directs its use and whether it forms part of a lawyer-led legal process.
What law firms need to know
For law firms, this is a pivotal moment in the evolution of legal AI.
Until now, much of the discussion around AI governance has focused on hallucinations, accuracy, cybersecurity and regulatory compliance. This decision elevates privilege and confidentiality to the centre of every AI conversation.
The ruling reinforces an important distinction between public consumer AI tools and enterprise AI platforms. Consumer services typically reserve broad rights over user inputs and outputs, while enterprise legal AI solutions often provide contractual commitments around confidentiality, data isolation and the non-use of customer information for model training. Those differences may prove increasingly important when courts assess whether communications remain confidential.
Are clients’ use of AI a legal risk?
It also changes how firms should think about clients’ use of AI before seeking legal advice.
Clients increasingly arrive at law firms having already used ChatGPT, Claude or similar tools to analyse contracts, draft responses to regulators or assess potential claims. Until now, many firms may have viewed this simply as another source of background information. Following this case, firms may instead need to consider whether clients have inadvertently compromised confidentiality before legal advice even begins.
The judgment also raises questions about internal firm policies. If lawyers encourage clients to use AI, what guidance should accompany that recommendation? Should firms prohibit clients from uploading confidential facts into public AI tools? Should engagement letters include warnings about generative AI? Should AI-assisted work only be undertaken within approved enterprise environments?
These are practical risk management issues that firms will increasingly need to address.
AI governance as a professional responsibility
The decision is also likely to accelerate investment in firm-approved AI environments. Rather than allowing lawyers to choose whichever public AI tool they prefer, many firms may move towards centrally governed AI platforms with contractual privacy protections, audit capabilities and clear policies on what information can and cannot be shared.
Importantly, this case should not be interpreted as a warning against using AI altogether. The judgment arguably points in the opposite direction.
Judge Rakoff acknowledged that AI used under the direction of counsel may have been viewed differently. That observation suggests the future of legal AI will not depend solely on technological capability, but on governance. Firms that can demonstrate appropriate supervision, clear policies, secure AI environments and well-defined legal workflows may be in a much stronger position than those relying on ad hoc use of public chatbots.
There is also reason to believe this will not be the final word on AI privilege. On the very same day as this ruling, another US federal court took a different view in a separate case, describing generative AI as a tool rather than a person and suggesting that AI-generated material could, in some circumstances, reflect a user’s own mental impressions. Legal commentators have also questioned whether the Heppner case gives too much weight to the existence of public AI platforms and too little to how AI functions within the lawyer-client relationship. As more cases emerge, courts may develop a more nuanced approach rather than adopting a blanket rule.
What should firms do now?
AI governance is becoming a core professional responsibility that touches confidentiality, privilege, client relationships and risk management. Firms should review their AI policies, ensure lawyers understand the distinction between public and enterprise AI systems, provide clear guidance to clients about using AI before seeking legal advice and consider whether existing confidentiality and engagement procedures remain fit for purpose.
The legal profession has spent the past two years asking how AI will change legal practice. This decision reframes the question. The real challenge is no longer simply how lawyers use AI, but how they use it without undermining one of the profession’s oldest and most fundamental protections.
