Why Using AI to Build an Immigration Case Can Backfire
Artificial intelligence can draft a petition letter in seconds. It can also invent a court case in the same breath. In October 2026, the Board of Immigration Appeals (BIA) showed what that mistake can cost, and a survey of court errors worldwide shows that even judges are not immune. This article explains why unverified AI use in an immigration profile or filing can hurt an applicant. The sources below concern lawyers and courts rather than applicants' own AI use. Here, our EB-1A consultants have dived in why nothing beats well positioned immigration cases by human expertise and experience.
The first BIA ruling on AI hallucinations
On October 1, 2026, the BIA published Matter of Sethi, 30 I&N Dec. 112, as precedent, according to Novo Legal Group. The Board suspended an immigration attorney for six months from practice before the BIA, the Immigration Courts and the Department of Homeland Security, retroactive to June 13, 2026. The suspension followed a Ninth Circuit suspension in Malkeet Lnu v. Blanche.
The facts are instructive. The attorney's brief cited cases that did not exist. When the problem surfaced, he filed a motion calling the fake citations "typographical errors." The Ninth Circuit found that implausible, since a typo cannot conjure a case that was never decided. He admitted he did not personally read the cases cited in briefs drafted by his firm's brief writers. He later conceded it was more likely than not that AI had been used.
He argued that discipline for AI-related conduct was unprecedented. The Board rejected that argument: the tool does not change the professional rules. Duties of candor and verification apply whether a brief is written by hand, delegated or generated by software.
What that means for the applicant
An applicant bears the consequences of a lawyer's shortcuts. A suspended attorney cannot practice before the agencies during the suspension, so a client could lose counsel mid-case. Applicants and petitioners also sign their own forms, and the signature certifies that the contents are true. If AI-written text slips an invented fact into a biography, a letter or a personal statement, the signer owns it.
Even judges are being fooled
Barrister Matthew Lee surveyed eleven matters in the U.S., India, the U.K. and South Africa in which false or AI-linked material entered a court's own decision. His count is non-exhaustive, and he distinguishes confirmed AI use from suspected use and from cases where the source was never established.
The consequences were real:
- In five matters, an appellate court reversed, vacated or set aside all or part of a decision.
- In Shahid v. Esaam, a lawyer was fined $2,500 after an order relied on fictitious cases.
- In Payne v. The State, the Supreme Court of Georgia vacated an order, partly because nine citations did not exist or did not support their propositions, and suspended the prosecutor's privilege to practice before it for six months.
- In a California matter, Campos, the court imposed $5,000 in sanctions and referred the opinion to the State Bar.
Lee's most unsettling finding is that origins can be mixed. In Campos, one fictitious authority traced to a Reddit post, while another appeared to be AI-generated. He concludes that verification aimed only at declared AI use is unlikely to catch every fabrication. He also warns of a cascade effect, in which errors repeated across documents begin to look credible. A South African judge put the principle plainly: "Judicial authority must remain human, accountable and transparent."
The machine reading your file
Applicants also face AI on the receiving end. Cozen O'Connor reports that USCIS uses an Evidence Classifier that tags uploaded documents and decides what adjudicators see first, a Verification Match Model that flags mismatched names and dates, and fraud-detection systems that look for anomalies. The firm links this to rising RFEs and cites "ghost text" detection, in which hidden text layers in PDFs trigger incorrect assertions. It also reports that EB-2 NIW denial rates have climbed to nearly 40 percent and that EB-1A petitions face heightened scrutiny.
If separate AI tools draft separate letters, small contradictions in titles, dates or project descriptions are likely. An automated cross-check is exactly what catches them.
Why generic text hurts EB-1A cases
Reddy Neumann Brown says adjudicators increasingly judge EB-1A reference letters by "the wrong test applied to the wrong category," treating them like employment-verification letters. That makes precision more important. Arvian Immigration says RFEs about letters most often stem from general wording, hyperbole without facts and weak links to exhibits. Immigration Copilot adds that expert letters should contain specific, verifiable claims directly linked to identified evidence. Language models tend toward polished generalities, and in an extraordinary-ability case, this kind of generality could be a fatal weakness.
A safer approach
Our EB-1A experts believe that a safer and more prudent approach in this case would be the following:
- Verify every citation, fact, date and quotation against a primary source before filing.
- Treat AI output as a first draft, never as evidence.
- Keep names, titles and dates consistent across all documents.
- Do not describe AI errors as "typos," and answer honestly if asked how a document was prepared.
- Ask an attorney how they use AI and verify filings. Novo Legal advises checking an attorney's standing with their state bar and asking about quality control.
At GCEB1, we are constantly discussing and dissecting the latest issues in the US immigration landscape. You can get in touch with us if you are looking for personalized merit based immigration strategies. We wish you a safe and stress free immigration journey ahead.
Frequently Asked Questions
1. What is an AI hallucination?
Content that sounds authoritative but is fabricated, such as a case citation or quotation that does not exist.
2. What was Matter of Sethi about?
The BIA suspended an attorney for six months after a Ninth Circuit finding that his brief cited nonexistent cases, and that he mischaracterized them as "typographical errors."
3. Does the ruling ban AI in immigration practice?
No. According to Novo Legal, it does not prohibit AI use. It addresses unverified output and false statements about it.
4. Can an applicant be harmed by their lawyer's AI mistake?
Potentially. Attorney discipline can interrupt representation, and filings with errors can draw scrutiny. That is my analysis, not a finding in the sources.
5. Do judges also make AI errors?
Lee's survey identified eleven matters with false or AI-linked material in judicial decisions, though in several the source of the fabrication was not established.
6. Does USCIS use AI on my petition?
Cozen O'Connor reports USCIS uses AI tools for evidence classification, data matching and fraud detection. Other sources say humans still make the final decisions.
7. Why are generic recommendation letters risky?
Sources say letters need specific, verifiable claims tied to evidence, and vague language is a common RFE trigger.
8. Can I use AI to organize my own documents?
Many people do, but every output needs human review. Consult an attorney about your situation.
9. What should I do if I find an error after filing?
Tell your attorney promptly so it can be corrected honestly. The Sethi attorney's disguised explanation was part of the problem.
Sources & further readings
- Arvian Immigration. "Employment-Based Immigration Recommendation Letters for EB-1 and NIW: How to Build Strong Letters in 2026." Accessed October 2, 2026. https://arvian-immigration.com/how-to-write-a-strong-letter-of-recommendation-for-an-eb-1-or-niw-petition/.
- Bettridge, Scott, and David S. Adams. "Growing Use of Artificial Intelligence in U.S. Immigration Adjudications Is Driving Higher RFE and Denial Rates." Cozen O'Connor, April 27, 2026. https://www.cozen.com/news-resources/publications/2026/growing-use-of-artificial-intelligence-in-u-s-immigration-adjudications-is-driving-higher-rfe-and-denial-rates.
- Cortes Romero, Luis. "Matter of Sethi Explained: BIA Suspends Attorney for AI Misconduct." Novo Legal Group, October 1, 2026. https://www.novo-legal.com/en/blog/bia-matter-sethi-ai-hallucinations-attorney-discipline.
- Immigration Copilot. "USCIS AI Adjudication: Preparing EB-1A Petitions in 2026." Accessed October 2, 2026. https://immigrationcopilot.com/blog/uscis-ai-adjudication-eb1a-2026.
- Lee, Matthew. "Judicial AI Hallucinations or Errors: 11 Cases and Counting." Natural and Artificial Law, August 15, 2026. https://naturalandartificiallaw.com/judicial-ai-hallucination-cases/.
- Reddy Neumann Brown PC. "Why Does USCIS Keep Misreading the Law? Current RFE Trends in EB-1A and EB-2 NIW Petitions." Accessed October 2, 2026. https://www.rnlawgroup.com/why-does-uscis-keep-misreading-the-law-current-rfe-trends-in-eb-1a-and-eb-2-niw-petitions/.
Perplexity
ChatGPT
Claude
Gemini









