Sullivan & Cromwell Admits AI Hallucinations in Client Work as Oregon Court Sanctions Lawyer for Fabricated Authorities
Two events this week crystallise the AI liability fault lines now running through legal practice. Sullivan & Cromwell has publicly admitted that generative AI hallucinations affected work product delivered to clients, marking one of the first acknowledgements of the problem by an elite Magic Circle-equivalent US firm. The admission lands the same day an Oregon state appellate court sanctioned a lawyer and ordered her to pay attorney fees after she filed a brief containing a fabricated list of authorities produced by generative AI — the first such sanction issued by that court. Taken together, the two episodes signal that AI-generated error is no longer a hypothetical risk management question for law firms and their insurers: it is live, documented, and attracting judicial and reputational consequence. The Oregon decision is notable because the court framed the filing as the attorney's professional responsibility regardless of the tool used, a formulation that closes off any delegation defence. For firms integrating AI into document drafting, due diligence, and brief preparation, both incidents provide hard data points for governance frameworks and professional indemnity exposure assessments.
Why this matters
Sullivan & Cromwell's admission is a watershed moment: elite firms have until now managed AI error through quiet remediation rather than public disclosure, and the reputational calculus of transparency versus concealment will now shift across the profession. The Oregon sanction reinforces that courts will not accept technological ignorance as mitigation — the professional duty of competence swallows the tool. Together, these events create reference points for professional indemnity underwriters, bar regulators, and general counsel evaluating whether to permit outside firms to use AI on sensitive matters. Law firms that have not yet formalised AI review protocols face heightened exposure both in court and under client engagement terms. The velocity of AI adoption in legal work makes this a systemic, not isolated, risk.
On the Ground
Trainees and associates reviewing AI-assisted drafts should treat verification of every cited authority as a non-delegable step, regardless of the platform or senior sign-off. Firms' risk and compliance functions are likely to circulate updated AI use policies referencing both incidents within weeks — watch for practice group guidance memos.
Interview prep
Soundbite
AI error is now a documented liability event, not a theoretical risk.
Question you might get
“Following the Sullivan & Cromwell hallucination admission and the Oregon sanction, how should law firms restructure their AI governance to manage professional liability?”
Full answer
The Sullivan & Cromwell admission and the Oregon sanction arrive almost simultaneously, which means regulators and clients can no longer treat AI hallucination as speculative. Firms need tiered review protocols: AI-generated citations must be independently verified against primary sources before any filing or client delivery, full stop. On the governance side, firms should designate a responsible supervising attorney for every AI-assisted work product, creating a clear audit trail. Professional indemnity policies should be reviewed to confirm coverage extends to AI-related errors, and client engagement letters may need to disclose AI use explicitly. The Oregon court's framing — that professional responsibility does not transfer to the tool — is the controlling legal principle every firm should build its policy around.
Sources
My notes
saved