EB-1A · Extraordinary Ability · Profession Guide

EB-1A for AI and ML researchers: Kazarian Criteria & AAO Patterns

How AI and ML researchers satisfy the Kazarian two-step analysis: which of the ten regulatory criteria are most accessible for this profession, and what final-merits evidence has cleared AAO scrutiny.

Based on 6,362 real USCIS AAO decisions · Last updated May 2026

Short answer

EB-1A requires AI and ML researchers to meet at least 3 of the 10 Kazarian regulatory criteria and then clear a final-merits analysis that the petitioner has sustained national or international acclaim. EB-1A denials for AI/ML researchers tend to fault sustained-acclaim: the evidence shows a strong post-PhD output burst but lacks multi-year continuity.

Most accessible Kazarian criteria for AI and ML researchers

The regulation requires that you meet at least 3 of 10 criteria from 8 CFR § 204.5(h)(3). Below are the criteria most commonly satisfied in EB-1A petitions by AI and ML researchers, with profession-specific evidence patterns.

  1. 1

    Original contributions of major significance

    Widely-cited papers with independent citation context (review articles, downstream methods built on yours), models or datasets with measurable adoption (HuggingFace download counts, benchmark leaderboard positions), or open-source frameworks with downstream users.

  2. 2

    Authorship of scholarly articles

    First/second-author papers at NeurIPS, ICML, ICLR, ACL, EMNLP, CVPR with high acceptance-rate evidence and independent citations.

  3. 3

    Service as a judge of others' work

    Program-committee membership (NeurIPS, ICLR, etc.), area-chair roles, peer review for top venues, journal review for TPAMI / JMLR.

  4. 4

    Original contributions — alternate framing via patents

    Patents on training methods, inference optimizations, or model architectures used in production systems.

Final-merits framing under Kazarian step 2

AI/ML is one of the more EB-1A-favorable fields when the petition can demonstrate sustained recognition (citation trajectory across multiple years, downstream methods named after the petitioner's work, invited talks at multiple venues). Final-merits denial usually reflects a single-paper / single-conference profile that doesn't demonstrate sustained acclaim.

Why EB-1A petitions by AI and ML researchers fail at AAO

EB-1A denials for AI/ML researchers tend to fault sustained-acclaim: the evidence shows a strong post-PhD output burst but lacks multi-year continuity. Several first-author papers across multiple venues over 3+ years, with rising citation counts, is the cleanest counter.

For context: across all professions, 5.9% of NIW appeals are approved at the AAO level. EB-1A appeals follow similar dynamics — most denials are at first-pass USCIS, and AAO data reveals which arguments fail at the highest scrutiny level.

Build your EB-1A petition with profession-specific framing

Our $99 EB-1A Petition Builder generates a Kazarian-framework petition letter section by section, with criterion-by-criterion evidence framing tailored to your profile and references to similar approved AAO cases in our 6,362-decision corpus.

One-time payment, no subscription. Greenway AI is a data + document-generation platform, not a law firm; nothing here is legal advice.