AI Likely Behind Reopening of Old Green Card Cases

RFEs, NOIDs and citizenship interviews now test old filings against each other with a precision not seen before. Public records show why.

U.S. Citizenship and Immigration Services appears to be relying more and more on AI tools to review applications. The clearest sign is in the cases themselves. Approvals granted years ago are being questioned for accuracy, and new requests list dates and job details side by side against applications filed long before.

Officers have always had the full file. What has changed is that the data in it is now cross-checked, line by line, across every earlier filing.

The cases coming up

I-485 RFEs and NOIDs. Employment-based applicants are receiving requests for evidence and notices of intent to deny that compare the work experience in their PERM labor certification with the employment history on other forms, including past DS-160 visa applications and the I-485 itself. When a title, employer or date does not match, the applicant must explain. Some NOIDs go further. They treat any untrue employment history in the PERM as a false statement made under penalty of perjury, even if the detail played no part in the approval, and tell the applicant to file a Form I-601 waiver. By law, a misrepresentation bars a green card only if it was willful and material. The waiver also requires a U.S. citizen or permanent resident spouse or parent who would suffer extreme hardship, and many employment-based applicants have no such relative.

N-400s that reopen the green card. At naturalization interviews, officers are going back to the I-485 approval. Applicants have been asked about a vaccination missing from the medical exam, a required affidavit of support that was never filed, and employment dates that differ across the PERM, I-140, I-485 and visa applications. In many of these cases, the officer who approved the green card had the same file and raised nothing.

That earlier mistake can still cost the applicant. To naturalize, a person must have been “lawfully admitted for permanent residence.” Under Matter of Koloamatangi (BIA 2003), a green card granted in error does not count, even if the government made the error and there was no fraud. In Turfah v. USCIS (6th Cir. 2017), the Sixth Circuit upheld a citizenship denial on that basis more than 20 years after the man entered. A denied N-400 can be followed by a Notice to Appear in immigration court.

Voter rolls. Applicants are also being told at interviews that they are registered to vote, often through a DMV form they did not understand. DHS has been matching applicants against state voter records. Our unlawful voting FAQ explains the consequences.

What USCIS’s own records show

USCIS does not tell applicants when software has flagged a case. But its public records point in one direction. In March 2026 the agency said its earlier screening and vetting had been “wholly inadequate.” Director Joseph Edlow has said USCIS is re-vetting people who were already granted green cards and other benefits.

DHS’s public AI inventory lists USCIS tools that:

match a person’s records across agency systems and link them into a single identity view;
suggest aliases and birth dates from existing records for officers reviewing N-400s;
flag similar narrative text across different applications;
tag uploaded evidence by document type; and
translate foreign documents with generative AI, including during interviews, a tool DHS last listed as in testing.

Behind these sits ATLAS, USCIS’s automated screening system. DHS says screening results are kept for 100 years from the applicant’s date of birth, so nothing in the file ages out.

The companies involved

Thomson Reuters. In August 2026, USCIS sought a vendor for “enhanced vetting activities involving individuals throughout the USCIS immigration lifecycle,” including batch processing of target records and identity matching against commercial and public data. It awarded a $2.8 million pilot, without competition, to Thomson Reuters Special Services.
Palantir. Since October 2025, USCIS has paid Palantir for successive phases of a platform for vetting marriage-based cases, described in contract records as a move to a “proactive, technology-driven approach.” In February 2026, DHS signed a department-wide deal worth up to $1 billion that lets USCIS and other agencies buy Palantir software without new bidding.
What comes next. A USCIS forecast for its main systems-integration contract, due to be awarded in 2027, lists “AI-enabled anomaly detection” and “risk scoring” among its goals.

USCIS says contractor tools support analysis and do not make eligibility decisions. A human officer still decides. But that officer now starts with every conflict in the file already laid out.

What applicants should do

Review the old filings, not just the new one. Compare names, dates, employers, job titles, addresses and marital history across every form ever filed. A USCIS FOIA request for the A-file usually returns past petitions and applications. Old DS-160s are harder to get, because visa records are confidential by law, so keep your own copies.
Use AI yourself. The same kind of tools USCIS relies on can find conflicts across dozens of documents, and check old translations, far faster than a person reading page by page. Use a secure tool, preferably with your attorney, because these files hold sensitive personal data.
Correct mistakes in the next filing. If an earlier date was a good-faith guess that turned out wrong, give the right answer with a short explanation and supporting records. Do not repeat the old answer. A correction made voluntarily, before USCIS raises it, can cure a misstatement under the BIA’s timely retraction doctrine. One made after the officer points it out does not.
Think before filing the N-400. If the review turns up a problem with how the green card was approved, a citizenship application can put the green card itself at risk. Get advice on whether to fix the problem first or wait.

This article draws on accounts from applicants and immigration attorneys over the past several months. The author is an immigration attorney. This article is for general information and is not legal advice.

Sources

Vasanth Vaidyanathan, Esq.

Administrator

Vasanth Vaidyanathan is an immigration attorney with over 20 years of experience focused exclusively on U.S. immigration and nationality law. He holds an LL.M. from California Western School of Law, San Diego and advises individuals and businesses on employment and family-based immigration matters, including H-1B, PERM, EB categories, and adjustment of status. He has also been a featured speaker at immigration seminars across the U.S. To reach Vasanth visit: https://www.vnathan.com/

Author's website Author's posts