Garcia Perez v. Blanche

What did the court warn about?

Judge Robert Pitman imposed no sanction. In footnote 4 of an August 18, 2026 order, the court stated that "Petitioner's counsel is strongly cautioned that future inaccurate citations 'hallucinated' by generative artificial intelligence tools may result in sanctions," citing Fed. R. Civ. P. 11(b). The court did not let the citations decide the merits: it said it "will nonetheless consider Petitioner's due process argument, as it will not punish Petitioner for his counsel's mistakes." Relying on real Fifth Circuit and Supreme Court authority that § 1226(e) does not bar review of constitutional challenges to detention, it went on to the due-process question and observed that "it appears possible that the immigration judge relied on inaccurate information in choosing to deny Petitioner bond." It ordered Respondents to file supplemental briefing by August 25, 2026 on whether the immigration judge incorrectly found Petitioner not immediately eligible to adjust status and, if so, whether that finding amounts to a due process violation, and allowed Petitioner to respond by September 1, 2026. The order does not decide the petition.

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

Garcia Perez v. Blanche is a habeas matter in which the court’s handling of fabricated authority sits in a single footnote, but the footnote is specific. Judge Pitman found that “an alarming number of cases cited by Petitioner for support in his Reply are nonexistent, ‘hallucinated’ cases,” listed five of them with their purported docket numbers and Westlaw cites, and reported that “[n]one of these cases exist on Westlaw or on the purported courts’ dockets.” The Reply also cited “Garcia v. Mullin,” a decision carrying the same “RP” docket suffix as this case and “purportedly written by this Court.”

The court separated the client from the lawyer. It stated that it “will nonetheless consider Petitioner’s due process argument, as it will not punish Petitioner for his counsel’s mistakes,” and then worked through the jurisdictional question using real Fifth Circuit and Supreme Court authority, ultimately ordering supplemental briefing on whether the immigration judge relied on inaccurate information when denying bond. The consequence for counsel was a caution, not a sanction: “Petitioner’s counsel is strongly cautioned that future inaccurate citations ‘hallucinated’ by generative artificial intelligence tools may result in sanctions,” with a citation to Rule 11(b).

The order does not make a finding that counsel used AI and does not name the attorney; its reference to generative AI appears in the forward-looking warning. Firms documenting compliance may wish to consider that here the court verified the citations against both a commercial database and the purported courts’ own dockets, so a citation that looks formally complete (docket number, WL number, pin cite, date) did not survive the check.