A Legal AI Tool Made Up Case Quotes and Derailed a Criminal Appeal

What happened

The appeal in United States v. Farris was moving through the standard stages of post-conviction review when the process stopped abruptly. Court-appointed attorney Steven N. Howe had submitted appellate briefs that included quotations attributed to real cases that did not say what the briefs claimed, and descriptions of precedent that did not match the actual rulings. Howe later admitted he had used Westlaw CoCounsel, a generative AI legal research tool, to draft the filings.

The problem with the briefs was not that they leaned on strained interpretations of existing law. The quotations themselves were false. The AI assistant produced text attributed to real cases that those cases did not contain, a category of error commonly called hallucination. In addition to the fabricated quotes, the briefs included misleading descriptions of what the cited precedent actually held. Howe did not catch either category of error before filing them with the court.

The court responded by appointing new counsel to replace Howe and restarting the appellate briefing process entirely. That reset imposed a direct delay on the defendant, whose case was interrupted through no fault of his own. Howe lost the compensation he had been appointed to receive for the work and faced the possibility of disciplinary action.

The Farris case arrives in a landscape already shaped by prior incidents of the same kind. A high-profile 2023 case, Mata v. Avianca, drew national attention when attorneys filed AI-generated briefs containing invented citations, prompting bar associations and courts to issue guidance requiring lawyers to certify that AI-assisted filings had been independently checked. The Farris filing suggests that certification requirements alone are not enough to prevent the problem. Howe's signature on the briefs represented exactly that kind of personal attestation, and it did not stop the fabricated quotations from reaching the court.

What the Farris case makes concrete is the absence of a documented verification step between what the AI tool outputs and what goes into the filed brief. Legal workflow places that responsibility on the individual attorney, but it does so through professional norms rather than an enforced process with a trail. When something goes wrong, there is no record of what the tool produced, what the attorney reviewed, and whether the two were ever compared. A provable record of what a system did and what a person checked before signing would make that gap auditable, and would surface the omission before a client's appeal has to start over.

Reported impact

Affected parties

Not publicly disclosed

Harm type

Not publicly disclosed

Scale

Not publicly disclosed

Financial impact

Not publicly disclosed

Regulatory action

Not publicly disclosed

Classification

Organization

Not publicly disclosed

AI system

Not publicly disclosed

Industry

Not publicly disclosed

Country

Not publicly disclosed

Provider

Not publicly disclosed

Incident type

Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

• No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.