Courts sanction lawyers for AI-hallucinated case citations and a federal ruling finds that using public AI chatbots may waive attorney-client privilege
A Mondaq analysis published this week consolidates two distinct but related legal risk developments that are now active in US courts and increasingly relevant to UK and global practice. First, courts continue to sanction lawyers for submitting filings that cite non-existent cases generated by AI tools, a phenomenon known as 'hallucination', where a large language model (LLM) produces plausible-sounding but fabricated case references. Sanctions have been imposed by multiple courts, and the pattern is sufficiently established that bar associations and court systems are now issuing specific guidance on AI-assisted legal research. Second, and more structurally significant, a recent US federal court decision found that a lawyer's use of a public AI chatbot to analyse or process client information may constitute a waiver of attorney-client privilege, the fundamental legal protection that keeps communications between a lawyer and client confidential. The logic is that submitting privileged information to a third-party AI service, particularly a consumer-grade public chatbot without enterprise-grade data handling guarantees, could be treated as a voluntary disclosure to a third party, which traditionally breaks the privilege. Together, these two developments create a clear dual liability for legal professionals using AI tools without adequate governance: sanctions risk from hallucinated output, and privilege waiver risk from uncontrolled data input. For City firms advising clients, the question of which AI tools are approved for use and what client information can permissibly flow through them is now a matter of professional conduct, not just operational preference.
Why this matters
The privilege waiver finding is the more significant of the two developments for UK practice, because attorney-client privilege (called legal professional privilege in English law) is a foundational protection that clients rely on when disclosing sensitive information to their lawyers. If using a public LLM to process that information breaks privilege, the consequences for a client in litigation could be severe, since the opposing party could potentially access communications that would otherwise be protected. UK courts have not yet ruled directly on this point, but the US federal court reasoning is likely to influence how English courts would approach the same question. Firms are already moving to restrict use of public AI tools on client matters and to require enterprise-grade data handling agreements with AI vendors. This creates immediate demand for AI governance policy drafting, vendor due diligence, and data processing agreement review.
On the Ground
A trainee on an AI governance matter would assist with drafting or reviewing data processing agreements with AI tool vendors, preparing a vendor due diligence questionnaire covering data handling, privilege protections, and model training practices, and helping to draft an internal AI governance policy for the firm's professional responsibility committee.
Interview prep
Question you might get
“How might a lawyer's use of a public AI chatbot to analyse client documents affect legal professional privilege, and what steps should firms take to manage that risk?”
Sign up free to see the full answer
A model answer you can lift into an interview — how to frame this story for a partner.
Sign up freeMy notes
saved