AI Generated Image
The rules for AI in immigration court are now part of a government-wide review. The Administrative Conference of the United States, the independent federal agency that advises agencies on procedure, asked the public on Nov. 19 how adjudicators, lawyers and unrepresented parties should be allowed to use generative tools in agency hearings, and it singled out the Executive Office for Immigration Review as one of two agencies that already have a written policy.
The request for comments, published at 91 FR 74534 (Doc. 2026-23539), is open until 10 a.m. Eastern on Jan. 19, 2027. It uses an unfamiliar term, “Generative Super Intelligence,” or GenSI. A footnote explains why: Executive Order 14434, published in October, directs agencies to say “Super Intelligence” in place of “Artificial Intelligence” in nonstatutory documents. The subject is the same technology lawyers already call generative AI.
Why it lands now. Immigration practice is about to get faster and less forgiving. The Board of Immigration Appeals’ final appellate rule cuts the appeal deadline to 10 days starting Dec. 17, makes summary dismissal the default and waives arguments left out of the notice of appeal. This week the Board also published Matter of K-L-S-, warning that speculative appeals can draw sanctions. Compressed deadlines are exactly where drafting tools are most tempting and least checked.
What ACUS is asking about AI in agency hearings
The notice poses 10 questions. They cover how participants use the tools now; whether adjudicators and agency staff should use them to help decide cases; whether existing duties of competence, confidentiality and candor already cover the problem; whether agencies need their own rules, benchbooks or standing orders; whether parties should have to disclose or certify AI use; what sanctions should follow misuse; and whether lawyers, nonlawyer representatives, self-represented parties and agency counsel should face different rules.
ACUS frames the risks plainly. The tools “can hallucinate,” generating false information; they open the door to deepfake evidence; and nonsecure tools raise privacy concerns. It also notes they “may be especially useful for self-represented parties,” a point with obvious weight in immigration court, where many respondents appear without counsel. Comments go by email to lrobbins@acus.gov, with the notice title in the subject line, or by mail to ACUS in Washington, and will ordinarily be posted on the project page.
The EOIR policy ACUS points to
The notice cites EOIR Policy Memorandum 25-40, issued Aug. 8, 2025, by Acting Director Sirce Owen, alongside a Labor Department administrative law judge standing order and Patent and Trademark Office guidance. The EOIR memo is permissive on its face. EOIR has “neither a blanket prohibition on the use of generative AI in its proceedings” nor a mandatory disclosure rule. The teeth are in the discipline rules: a practitioner who submits hallucinated or erroneous AI-generated content likely violates professional obligations and can be disciplined under 8 C.F.R. 1003.101 and 1003.102. The memo “strongly” cautions judges about their own use and lets individual judges issue standing orders with management approval.
That approach has already been tested. The Board in October suspended a lawyer over AI-invented citations, and federal appeals courts have moved the same way. A Ninth Circuit opinion in Lnu v. Blanche (June 3, 2026) fined two attorneys, suspended them from practice before that court for six months and required AI disclosure statements for two years. The court put the principle in one sentence: “the rules are not violated at the point of research and drafting, but at the point of signing and filing.”
The government side uses the technology too. The DHS AI use case inventory for USCIS lists deployed tools such as an evidence classifier that tags pages in electronic case files for officers, and a sentence-similarity text analytics tool that DHS labels rights-impacting. ACUS’s questions about agency staff reach that side of the table as well.
Where this is likely to lead
ACUS cannot bind EOIR. Its usual product is a recommendation adopted by its assembly, which agencies may follow or ignore. It has addressed the technology before, in a 2020 statement on agency use of artificial intelligence. The questions posed this time point toward disclosure or certification duties and clear sanctions rather than a ban.
For immigration practitioners, the practical stakes are twofold. A recommendation favoring certification would give EOIR a template for converting PM 25-40’s interim guidance into a firmer rule, likely through a policy manual change or standing orders rather than formal rulemaking. And comments describing how pro se respondents use these tools could shape whether any rule distinguishes them from lawyers, the tenth question on the list. With EOIR’s corps of more than 800 immigration judges handling a growing docket, how judges themselves may use the tools may prove the more consequential answer.
This article is for general information and is not legal advice.
Sources
- ACUS, “Policies on the Use of Generative Super Intelligence in Agency Adjudication; Request for Comments,” 91 FR 74534 (Nov. 19, 2026)
- ACUS project page
- EOIR Policy Memorandum 25-40, Use of Generative Artificial Intelligence in EOIR Proceedings (Aug. 8, 2025)
- Lnu v. Blanche, No. 24-4790 (9th Cir. June 3, 2026)
- DHS AI Use Case Inventory, USCIS
- ACUS Statement #20, Agency Use of Artificial Intelligence (2020)