What is the role of Gen AI In Litigation? AI, Legal Advice and Privilege

What happened in the US v Heppner case?

In US v Heppner, the defendant created 31 documents using output generated by Claude on potential defence strategy and legal arguments. The defendant later shared these AI generated documents with his lawyers and asserted attorney client privilege and protection under the work-product doctrine over them under US privilege principles.

Under US law ‘attorney-client privilege’ (communications between lawyer and client to give and receive legal advice) or the ‘work product doctrine’ (which provides protection for material prepared by or for counsel in anticipation of litigation) can enable a litigant to withhold documents from disclosure.

The Federal Court ruled that the defendant's interactions with Claude were not privileged and were therefore disclosable in the court proceedings. Concluding that the documents failed at every stage of the privilege analysis, the Federal Court’s reasons for the decision were:

1. There was no communication with a lawyer – Claude, and AI tools generally, do not have legal personality and as such are not a lawyer, nor able to enter into an attorney client relationship. The court stressed that using an AI tool to analyse legal issues is “no different from discussing your case with a friend.”
2. Lack of confidentiality – Claude’s privacy policy expressly permitted the platform to collect user prompts and outputs and use them to train the model. Communications with the tool were therefore not confidential and could not meet the confidentiality limb of privilege.
3. The documents were not created for the purpose of obtaining legal advice from a lawyer – The documents were prepared on the defendant’s own initiative, without direction from lawyers. In addition, the platform itself stated that it did not provide legal advice and was designed to avoid responses that could be characterised as legal advice.
4. The documents did not qualify for protection under the work product doctrine – Even if the material generated by Claude was prepared "in anticipation of litigation," it was not prepared by or at the request of the defendant's lawyer, a prerequisite for the doctrine to apply.

Same day, different outcome

Conversely on the same day, in Warner v. Gilbarco, Inc., No. 2:24-cv-12333, 2026 WL 373043 (E.D. Mich. Feb. 10, 2026), the US Eastern District Court of Michigan found that the use of AI tools did not constitute a waiver of work-product privilege. The defendant argued that the claimant waived the work-product protection by using ChatGPT, however the court rejected this argument on the basis that the work-product waiver has to be a waiver to “an adversary or in a way likely to get in an adversary's hand”.

The court noted that ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background. The defendants' motion asked the court to compel the claimant’s internal analysis and mental impressions (being her thought process), rather than any existing document or evidence, which was not discoverable as a matter of law. The court concluded that the motion sought intrusive post-discovery production based on speculation about what might exist in the claimant’s internal drafting process, and was ultimately considered a fishing expedition.

These decisions highlight the fact-specific nature of privilege analysis in the AI context. It is notable that the courts reached different conclusions, although two factual distinctions appear to be that in Warner (a) the claimant was acting as her own legal counsel in civil proceedings, and (b) that ChatGPT could not be a third party to which disclosure had been made. Ultimately, the decisions highlight the factual nuances that can impact whether privilege protection applies.

Would the courts in England and Wales reach the same conclusion under English Law?

While both cases were determined under US federal law, they offer valuable insight for all litigants. The principles of legal professional privilege under English law are different to the position in the US. The English courts may have reached the same conclusions in these circumstances, but with some differences in the reasoning.

Legal advice privilege

Under English law, legal advice privilege protects confidential communications between a lawyer (acting in a professional capacity) and their client (which is narrowly defined) made for the dominant purpose of giving or seeking legal advice.  AI tools, whether generative models or natural language assistants, do not have legal personality. Even if an AI tool provides output that resembles legal analysis, the absence of a lawyer on the “other side” of the communication is fatal. Legal advice privilege cannot attach to communications with non lawyers merely because the subject matter is legal. For the same reasons given in US v Heppner, no legal advice privilege arises where there is no presence of a lawyer and where the communications lose their confidentiality.

Litigation privilege

Litigation privilege covers communications (which includes documents) which are confidential, made by or between a client or its lawyer and a third party, and made for the sole or dominant purpose of obtaining information or advice in connection with existing or contemplated litigation. Litigation privilege is broader in scope than the US equivalent because it does not require the involvement of a lawyer to arise. The communications in US v Heppner and Warner may therefore have satisfied the dominant purpose test for litigation privilege.

Despite this, the documents may still have fallen foul of the requirement of maintaining confidentiality due to the use of public AI tools. However, the commentary from the US court as to whether use of an AI tool constitutes disclosure to a third party (therefore losing its confidentiality) is an interesting one and may remain open to further legal debate going forward.

Confidentiality remains a key challenge

Any disclosure to a third party, whether intentional or accidental, risks stripping a communication of the confidentiality required to sustain a claim to privilege. Each stage at which an AI tool handles privileged material therefore represents a potential vulnerability. That the processing is carried out by software rather than a human does not remove the legal risk; it merely obscures it and makes any resulting loss of privilege more difficult to identify or remedy.

Publicly accessible AI tools almost always permit the provider to access, store, or employ user inputs for model training or product improvement. Where the provider can access or use the user’s content, confidentiality cannot be maintained. Once confidentiality is lost, privilege is lost and therefore feeding privileged or sensitive information into a public AI platform would almost certainly constitute a waiver of privilege under English law and this appears to be the direction of travel in the English courts. For example:

• The October 2025 guidance to the judiciary on the use of AI states “Any information that you input into a public AI chatbot should be seen as being published to the world”.
• In a recent judgment the Immigration and Asylum Chamber of the Upper Tribunal gave a stark warning about putting confidential information into AI chatbots stating "Uploading confidential documents into an open-source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege".

The position in relation to private internal AI tools is less clear and organisations will need to give careful consideration to the potential risks of waiving privilege in how those systems are operated.

Key Takeaways

• Confidentiality is always a pre-requisite of legal privilege, without which any claim to legal privilege will fail
• Do not input privileged or sensitive information into public AI tools
• Privilege cannot attach to communications with AI platforms simply by the nature of legal questions or work being produced 
• AI-generated legal analysis is not necessarily protected by privilege, with significant risks that it is not
• AI tools are not classed as a lawyer
• Work with IT and procurement teams to ensure any AI tool used by the legal team and the business more widely is approved for complex business use, meets high levels of security, operates on zero-retention and no-training policy, is access restricted and subject to robust operation and confidentiality terms.

Read more here:

• Part 1: What is the role of Gen AI in litigation?