Sullivan & Cromwell apologises to court for AI-generated 'hallucinations' in filed brief, reigniting debate on law firm AI governance obligations
Sullivan & Cromwell has issued a formal apology to a court after submitting a filing containing inaccurate legal citations generated by an AI tool — the firm acknowledged that the errors stemmed from so-called 'hallucinations', where AI systems generate plausible-sounding but factually incorrect content. The firm confirmed it maintains comprehensive policies and training requirements governing AI use in legal work, but stated that those policies were not followed in this instance, and that a secondary review process also failed to identify the errors before the filing was submitted. Opposing counsel Boies Schiller Flexner flagged the inaccurate citations to the court — over 40 errors identified across a three-page list. The incident is the most high-profile AI filing error by a major firm since the Mata v Avianca sanctions case of 2023, and follows a pattern of courts responding to AI errors with heightened scrutiny of law firm AI governance. The letter to the court did not identify which AI programme was used. The episode has immediate practical implications: courts are increasingly treating AI citation errors as professional conduct issues rather than simple mistakes, and bar regulators in multiple jurisdictions are developing guidance on AI use in litigation. For UK-targeting students, the episode is directly relevant because the Solicitors Regulation Authority has signalled it is monitoring AI-related conduct risks, and major City firms are each developing their own AI governance frameworks against this backdrop.
Why this matters
The Sullivan & Cromwell hallucination incident activates the intersection of legal ethics, AI governance, and professional liability — three practice areas that City firms are simultaneously trying to develop. For law firm management, the episode is a governance failure that demonstrates AI tool risk does not disappear with a written policy; it requires active workflow integration and mandatory verification steps. The 'why now' trigger is the maturation of AI tool deployment at major firms: the first wave adopted AI cautiously and with heavy oversight, but as AI becomes embedded in routine legal drafting, the pressure to use it at speed increases the risk of policy non-compliance. For clients, the commercial implication is that any AI-assisted work product carries residual verification risk, which means fee arrangements and engagement letters need to address AI tool use explicitly. For UK students, the relevance is direct — the SRA's ongoing monitoring means City firms are accelerating their AI governance policy development, generating training and compliance work across all practice groups.
On the Ground
A trainee working on AI governance matters arising from this type of incident would assist with AI governance policy drafting for the firm's internal use, prepare vendor due diligence questionnaires for AI tool providers querying their hallucination rates and source citation reliability, and draft regulatory impact assessment memos mapping the SRA's published guidance on AI conduct risks against the firm's current tool deployment.
Interview prep
Soundbite
AI hallucination liability is no longer a tech risk — it is a professional conduct risk that sits squarely with the supervising partner.
Question you might get
“If a partner at your firm discovers that an AI tool produced a hallucinated case citation in a brief already filed with the court, what steps should the firm take and what professional conduct obligations are engaged?”
Full answer
Sullivan & Cromwell has apologised to a court after submitting a filing with AI-generated inaccurate citations, confirming that the firm's own AI policies were not followed and that its secondary review process failed. The incident matters beyond the embarrassment: courts are beginning to treat AI errors as professional responsibility issues, meaning supervising partners — not just the associates who drafted the filing — face potential sanctions exposure. For City firms, this accelerates the timeline for embedding AI verification into workflow rather than treating it as an optional additional step. The broader trend is a convergence between AI governance and professional conduct, which means every UK-qualified lawyer needs to understand not just how to use AI tools but how to document their verification process — the SRA's monitoring means this is moving from best practice to obligation.
Sources
My notes
saved