Morgan & Morgan AI Sanctions: How Internal Review Missed Fake Citations
The Mata v. Avianca matter introduced the bar to the hallucinated citation. The Morgan & Morgan matter introduced the bar to the hallucinated citation at scale. The difference is the lesson.
A solo practitioner missing a fabricated cite is a story about one attorney's workflow. A national plaintiffs' firm missing a fabricated cite is a story about how internal review works at every firm. The mechanism of the miss is not idiosyncratic. The mechanism is structural.
Why this matter is different from Mata
The attorneys in Mata v. Avianca worked in a small firm. The brief passed from one attorney to another. The verification step that did not occur could have been performed by a single person reading the reporters. The fact pattern fit a story the profession could put away: an individual workflow failure in a small shop.
Morgan & Morgan is the largest plaintiffs' firm in the United States. The firm has thousands of attorneys, a large litigation infrastructure, and the resources to maintain a layered review process. Press coverage in early 2025 reported that attorneys at the firm filed a brief containing AI-generated citations that did not resolve to real opinions. The brief was filed. The cites reached the docket.
The size of the firm is the entire point. If internal review at a firm of that scale does not catch an AI-generated citation, the inference is not that the firm is unusual. The inference is that internal review, as practiced, is not the right instrument for this failure mode.
The pattern, at the level public reporting establishes
The publicly reported pattern is straightforward. A draft was prepared with the assistance of a generative AI tool. The draft included citations the tool produced. The draft passed through the firm's internal review. The brief was filed. The court identified citations that did not resolve to real opinions. The firm responded with corrective filings and, subsequently, with firm-wide guidance to attorneys on AI use.
The granular details of which citations were fabricated, what sanctions issued, and which judges presided are matters of public record covered in the press at the time. The pattern is what matters here. A draft entered review. The review did not surface the cites. The draft became a filing.
This sequence has now occurred at firms across the size spectrum. The Morgan & Morgan matter is significant because it forecloses the argument that the failure mode lives only in small or under-resourced firms.
Why internal review at a large firm still fails for AI citations
Internal review at a firm is built on recognition. The partner reads the brief. The partner recognizes the cited authority. The partner notices when a cite reads off because the holding does not match the partner's memory of the case. The method works when the partner has read the cited cases in the last few years and when the cases are the kind of authority the partner sees often.
Recognition is not verification. Recognition is a probabilistic judgment that the cite looks like a real case. A fabricated citation in a real reporter format, naming a real circuit, in a real year, will not trigger recognition failure. The cite reads like other cites. The partner moves on.
The partner reviewing a thirty-page brief with thirty citations cannot recall whether every cited case actually exists. The memory load is too high, and the task is the wrong shape for memory. Verification is a lookup task. Recognition is a pattern task. A brief that passes the pattern task can still fail the lookup task.
That is the structural answer to the question of why a firm of this size missed the cites. The firm did not skip review. The firm ran review. Review, as the profession has practiced it for decades, was not built to catch this particular failure.
The firm-wide guidance response
After the matter became public, the firm issued internal guidance to attorneys on generative AI use. Other firms across the industry issued similar memos across 2025. The memos are useful and track ABA Formal Opinion 512. The memos define when AI may be used, who is responsible for the output, and what disclosures are required under Model Rule 1.1. The memos set policy.
A memo does not verify a citation. A memo describes the standard of care. The standard of care is met by the work performed on each filing, not by the policy that names the work. The distance between policy and practice is what produces the next sanction order.
The operational answer to the policy is a verification step that runs per filing. The step is short. The step produces a record. The record is what shows the policy was followed in the particular matter. The record is what the malpractice carrier reads, what the court reads if it asks, and what the firm carries in the matter file.
What a scanner returns on this kind of brief
We did exactly that with the citations from the Wadsworth motions in limine. The plaintiffs’ brief cited nine cases. Each was run, as it appeared, against CourtListener’s citation corpus. The output is a structured list with a verdict on each.
Eight of the nine do not resolve to any opinion. The ninth — United States v. Caraway — resolves cleanly: a real case, confirmed in the reporter. That is the line the firm’s internal review could not draw by reading. The reviewer does not need to read the brief to see that eight cites need work before filing. The job changes shape: from finding problems by reading every line to confirming resolutions on a structured list.
That is the shift that matters. The scanner does not exercise legal judgment. The scanner does not write the brief. The scanner performs the lookup task that recognition was never built to perform.
Tooling, not memos
The takeaway from the Morgan & Morgan matter is not that AI is dangerous. The takeaway is that internal review, on its own, is not the audit layer the profession needs for AI-assisted drafting. A firm that has issued a memo has set the policy. A firm that has run a scanner on each filing has executed the policy and produced the record.
The pre-filing citation check is the step that closes the loop between the policy in the memo and the brief on the docket. The Verification Certificate is the artifact that proves the step ran. The two together are the operational answer to the question that the Morgan & Morgan matter put to the bar.
Partner review continues. Partner judgment continues. The recognition task continues to do what recognition does well. The lookup task moves to a tool built for the lookup task. The division of labor is the point.