October 18, 2026 (in 16 days): Tenth Circuit: comments close on 2027 rules, incl. proposed Rule 46.5(B)(4) generative-AI review →

Perez-Castillo v. Blanche

U.S. Court of Appeals for the Seventh Circuit · 7th Cir. · Illinois bar guidance , Indiana bar guidance , Wisconsin bar guidance

Conduct

Counsel filed an outsourced brief he admitted he had not reviewed; it carried 24 false quotes, 7 bad cases, and false record facts. The drafter blamed ChatGPT.

Consequence

$5,000 FRAP 46(c) fine on signing counsel, drafter admonished, ARDC referral; show-cause stays open pending ARDC review, with status reports due.

Lesson

The signing lawyer answered for an outsourced brief despite saying it was not on him to redo it, and refiling without review compounded it.

Court sanction

Verified September 28, 2026

Citation
Perez-Castillo v. Blanche, No. 25-1988 (7th Cir. June 1, 2026) (Brennan, C.J.) (published), ECF No. 45
Decided
June 1, 2026

Summary

Heriberto Perez-Castillo petitioned for review of a Board of Immigration Appeals removal order. His counsel of record, Abdullah Salah, had outsourced the research and writing of the opening brief to Farah Chalisa, an Illinois lawyer who did not enter an appearance. The Court found the brief "riddled with AI hallucinations": twenty-four unattributable or outright false quotations, at least seven mislabeled or nonexistent cases (many presented as Seventh Circuit precedent), and factual assertions contradicted by the record, including a claim that the petitioner's children were "deeply integrated into local schools" when all were over twenty-five. Salah told the Court that he "had no input in the research and preparation of the brief," admitted at oral argument that he had not reviewed the original or the revised brief, and said he did not believe "it was on me to redo the brief." His revised brief, filed the night before argument, replaced the false citations but still contained unsupported assertions and at least two hallucinated quotations. Chalisa said she used ChatGPT only for copyediting and that it hallucinated facts and altered her correct citations; the Court found it "unlikely that such pervasive errors would arise through a grammar check."

AI tool:
ChatGPT, per the drafting attorney (Farah Chalisa said she "used ChatGPT to perform the copyediting tasks necessary" and not for legal research, while hinting at using "any system incorporated into Lexis/Nexis"; the Court found the brief "riddled with AI hallucinations" and doubted her grammar-check account)
Sanction amount:
$5,000 fine against attorney Abdullah Salah under Fed. R. App. P. 46(c); attorney Farah Chalisa admonished; matter referred to the Illinois Attorney Registration and Disciplinary Commission; show-cause order kept open pending ARDC review; no sanction on Chalisa "at this time"
This case summary is informational only. Verify the underlying opinion or order against the primary source before relying on it in any filing or client matter.

What sanction did the court impose?

In a published opinion by Chief Judge Brennan, the Court denied the petition for review on the merits and imposed a $5,000 fine on Salah under Federal Rule of Appellate Procedure 46(c) because he "failed to check the AI-hallucinated citations, legal propositions, and factual representations in two briefs he signed and submitted to this court." It held both versions of the brief violated FRAP 28 and declined to bar Salah from practice before the Court, warning that a further incident could lead to disbarment. Finding no evidence that Chalisa committed fraud on the court and nothing in her response indicating bad faith, it declined to sanction her "at this time" under its inherent authority, but admonished her. It referred the matter to the Illinois Attorney Registration and Disciplinary Commission, directing the clerk to send the opinion, both briefs, the attorneys' responses, and the argument recording "for such consideration as it sees fit," noting questions about Chalisa's account of her AI use and possible fee-splitting under Illinois Rules 1.5 and 1.2(e). The show-cause order stays open pending ARDC review; each lawyer must file a status report one year from the decision or within seven days of final ARDC action, whichever comes first, and "[a]t that point, we may consider whether additional sanctions are appropriate."

Why does Perez-Castillo v. Blanche matter for law firms using AI?

Perez-Castillo sorts responsibility between two lawyers with different roles. Salah signed the brief; Chalisa wrote it. The Court placed the sanction on the signature. Proceeding under Federal Rule of Appellate Procedure 46, which reaches “conduct unbecoming a member of the bar,” it held that signing and submitting a brief he admitted he had not reviewed “display[s] egregious misjudgment” and is inimical to the administration of justice. “What is more,” Salah filed the revised brief without review after the Court had put him on notice of its deficiencies. “Attorneys practicing before this court should never sign and submit a brief they have not reviewed.”

The opinion also explains where this case sits in the circuit’s recent practice. It contrasts Jones v. Kankakee County Sheriff’s Department, where a pro se litigant with a few nonexistent quotations was not sanctioned, and Dec v. Mullin, where an attorney whose erroneous standard-of-review citations were “presumably copied and pasted” from another brief, and who apologized, was only admonished. This case was “categorically different” because the brief combined false quotations, erroneous law, and misstated facts, and because the signing attorney “did not take responsibility or apologize.” Chalisa, by contrast, apologized and committed to AI-focused CLE; the Court found no evidence that she committed fraud on the court and nothing in her response indicating bad faith, and admonished her without sanction “at this time,” while doubting that “such pervasive errors would arise through a grammar check.”

The referral is the most unusual feature. The Court directed the clerk to send the ARDC the opinion, both briefs, the attorneys’ responses, and the argument recording “for such consideration as it sees fit.” It said an ARDC review would help it know whether Chalisa’s response accurately reflects her AI use, noted concern that the Salah-Chalisa arrangement raised the possibility of an impermissible fee split without the client’s written consent, and said the ARDC “may conduct a more searching review of these questions.” It kept its own show-cause order open, directing status reports one year from decision or within seven days of final ARDC action, whichever comes first, after which “we may consider whether additional sanctions are appropriate.” Firms documenting compliance may wish to consider that outsourcing brief-writing to another lawyer raises supervision, candor, and fee-sharing questions at once when AI is involved.

Implications for your firm

Operational steps a firm reading this case may wish to consider documenting. Strategic and rule-application calls belong to your firm's attorneys.

  • Review every brief before signing, including work contracted out to another lawyer. The Court told Salah that under the FRAP and its standards he was responsible for any brief going out in his name.
  • Document written client consent to any arrangement with outside counsel. The Court raised the possibility of an impermissible fee-splitting arrangement, noted that the Illinois Rules require written client consent (citing Rules 1.5(b), (f) and 1.2(e)), and said the ARDC 'may conduct a more searching review' of these questions.
  • Verify record facts as well as citations after any AI-assisted editing. The drafter's account was that a grammar check introduced facts not in the record; the Court doubted that account and stated that attorneys must ensure their work product contains no false legal citations or factual errors.
  • Consider that time pressure did not excuse filing a corrected brief without review. Salah blamed the refusal to postpone argument; the Court answered that he had 'over three weeks' after being alerted and '[t]hat was plenty of time to identify and correct errors.'

Sources

Primary sources