EB-1A for AI Researchers: Which Criteria Industry and Academic ML Researchers Actually Meet
Papers and conference reviewing often cover two of the three EB-1A criteria an ML researcher needs, but final merits decides the case. USCIS approved 32.2% of decided EB-1A petitions in Apr-Jun 2026.
Part of our guide to EB-1A and EB-2 NIW.
EB-1A for AI researchers usually comes down to two questions. First, can you document at least three of the ten criteria in 8 CFR 204.5(h)(3)? Many ML research scientists can, through peer-reviewed papers, conference and journal reviewing, and either a critical role at a distinguished lab or a high salary. Second, does your whole record show sustained acclaim at the very top of the field? That second step, the "final merits" review, is where an AI petition is usually won or lost.
Status as of November 1, 2026:
- Final merits review: in effect. Agarwal v. Miller (D. Neb., 2026-09-25) upheld it; Mukherji v. Miller (D. Neb., 2026-01-28) had ordered an approval, and the government dropped its appeal on 2026-06-10. No appellate court has ruled, and USCIS still applies it (analysis of the split).
- PA-2026-05: in effect, challenged. Since 2026-08-05, for every USCIS benefit request, officers may deny without an RFE or NOID when required initial evidence is missing (USCIS policy alert). AAUP v. Mullin challenges it; no ruling yet.
- Latest data: USCIS approved 32.2% of decided EB-1A petitions in FY2026 Q3 (April to June 2026), against 55.3% for NIW. Both predate PA-2026-05.
- November 2026 Visa Bulletin: published. EB-1 Final Action Dates are India 2023-02-01, China 2023-08-01, rest of world Current.
What does EB-1A for AI researchers require?
EB-1A is an employment-based green card you can file for yourself, with no job offer and no labor certification. You must show either a one-time major internationally recognized award or at least three of ten regulatory criteria, then persuade the officer that your whole record places you among the small percentage at the very top of your field.
The USCIS EB-1 page confirms "No offer of employment or labor certification is required," though you need evidence that you will keep working in your area of expertise in the US (8 CFR 204.5(h)(5)). The USCIS Policy Manual, Volume 6, Part F, Chapter 2 (current as of 2026-10-27) prescribes "a two-step analysis" from Kazarian v. USCIS, 596 F.3d 1115 (9th Cir. 2010): step one checks each criterion; step two asks whether you are "one of that small percentage who have risen to the very top of the field of endeavor."
Which EB-1A criteria do ML researchers actually meet?
Authorship and judging, most often. Original contributions, a critical role and a high salary depend on the record. Media about you and qualifying memberships are less common, and the two arts criteria rarely apply.
EB-1A criteria mapped to typical ML records (as of 2026-11, from 8 CFR 204.5(h)(3) and the Policy Manual)
| Criterion | Industry ML researcher | Academic ML researcher | What officers look for, or discount |
|---|---|---|---|
| (i) Lesser awards | Best-paper awards | Best-paper and selective dissertation awards | "Certain doctoral dissertation awards" can count; one open only to a single employer, school or locality "may have little national or international recognition" |
| (ii) Membership | Uncommon | Elected fellow grades, later in a career | Membership gained by "the payment of a fee", a degree or years of experience does not count |
| (iii) Published material about you | Tech press discussing you and your work | Press coverage of a paper | Must be about you, "not just about the person's employer"; marketing excluded |
| (iv) Judging | Conference and workshop reviewing | Reviewing, area chair, thesis committees | Proof you did the reviewing; an invitation alone is not enough |
| (v) Original contributions of major significance | Methods or models adopted beyond your team; open-source others build on | Widely adopted methods, datasets, benchmarks | Work "cited at a level indicative of major significance in the field"; letters that "specifically describe the person's contribution"; funding, a patent or publication "will not necessarily establish, on its own" major significance |
| (vi) Authorship | Peer-reviewed papers, sometimes fewer for confidentiality | Proceedings and journal papers | Met at step one "regardless of the caliber"; caliber matters at step two |
| (viii) Leading or critical role | Tech lead on a named system at a well-known lab | Lead on a flagship project | A title "with appropriate matching duties" helps; for a critical role, "the person's performance in the role" decides; the organization needs a distinguished reputation |
| (ix) High salary | Often documentable with W-2s and offers | Usually hard to meet | High "relative to the compensation paid to others working in the field"; user-reported salary data "may not be a valid comparison" |
| (vii), (x) Arts criteria | Rarely applies | Rarely applies | Comparable evidence only if a criterion "does not readily apply to the person's occupation" |
Two Policy Manual details matter for STEM records. First, comparable evidence is allowed only when you show a criterion does not readily apply to your occupation, not because your evidence for it is thin; the Policy Manual's own example is an industry researcher for whom a major trade-show presentation can be "of comparable significance" to scholarly publication. Second, the final merits step lists positive factors that fit research careers: articles in "particularly highly-ranked journals", a citation rate that is high for the field (such as a high h-index), "unsolicited invitations to speak or present research at nationally or internationally recognized conferences", and being "named as an investigator, scientist, or researcher on a peer-reviewed and competitively-funded U.S. government grant."
Can an AI researcher get EB-1A without a major award?
Yes. The one-time achievement route is for awards on the scale of the Nobel Prize, the Policy Manual's own example, so for almost any ML researcher the realistic route is three of ten. A best-paper award helps under criterion (i) but is not required.
Do NeurIPS or ICML papers satisfy EB-1A?
Usually, for authorship. The Policy Manual says scholarly articles meet that criterion "regardless of the caliber" and describes them as generally peer reviewed, so published proceedings fit better than a preprint that was never reviewed.
Papers alone do not show acclaim. At final merits the officer asks what your papers changed, who built on them and how your record compares with your subfield. The final merits factor names "particularly highly-ranked journals"; because ML's top venues are conferences, independent evidence such as acceptance rates or venue rankings should show where yours stand (our reading). On a many-author paper, letters should explain your specific contribution.
Does reviewing for conferences count as judging?
Yes, if you can prove it. The Policy Manual requires evidence that you were invited and actually judged, so save completion records from OpenReview, CMT or the journal's system, not only invitations.
Step two asks whether the judging shows top-of-field standing. A reviewer slot at a venue that recruits broadly says less than an area chair, senior program committee or editorial board role. Our guide to judging and peer-review opportunities for engineers covers documenting these roles for EB-1A and O-1.
How many citations do I need for EB-1A?
There is no number. For original contributions, the Policy Manual looks for work "cited at a level indicative of major significance in the field." At final merits, a citation rate that is "high relative to others in the field," such as a high h-index, is a positive factor, and when a petition relies on citation tools, "a petitioner should explain its methodology and significance."
It does not mention self-citation, but because the comparison is with others in the field, practitioners separate self- and co-author citations from independent ones. A profile total also mixes preprint, workshop and journal versions of the same work. Our breakdown of the citation evidence USCIS discounts covers each trap.
Industry or academic: where each ML record is weakest
Industry researchers often clear salary and critical role, but their best work may be internal, with proof coming from the employer. The Policy Manual says "the relationship or affiliation between the beneficiary and the witness is also a factor the officer should consider," so adoption by other labs, outside press and citations from unrelated groups carry more independent weight than a manager's letter.
Academic researchers usually have stronger authorship, judging and citations, but rarely a standout salary. With a permanent research offer, compare the employer-filed route in EB-1A vs EB-1B for outstanding researchers. EB-1B's FY2026 Q3 approval rate was 93.0%, but it has a different standard and needs a qualifying employer offer, so the figure is not your EB-1A odds. If you are still weighing industry labs against cap-exempt academic or nonprofit roles, see our AI safety researcher H-1B sponsorship guide.
How does final merits judge sustained acclaim in a fast field?
Officers look for acclaim kept up over time. The Policy Manual says "the term sustained does not imply an age limit on the beneficiary," and that a beneficiary "may be very young or early in his or her career and still be able to show sustained acclaim." It expects recognition "well beyond the circle of their personal and professional acquaintances."
In AI that cuts both ways. A recent paper can gather citations fast, which helps; a record built on one paper with nothing comparable since invites the question of whether acclaim was sustained. And because letters of support "should not form the cornerstone of a successful claim," a petition carried by advisor and coauthor letters is exposed at step two.
EB-1A approval rates in 2026, read carefully
USCIS I-140 approval rates, FY2026, approvals / (approvals + denials) (as of 2026-11)
| Classification | Q1 (Oct-Dec 2025) | Q2 (Jan-Mar 2026) | Q3 (Apr-Jun 2026) |
|---|---|---|---|
| EB-1A | 47.5% | 41.8% | 32.2% |
| EB-2 NIW | 42.5% | 48.1% | 55.3% |
Source: USCIS I-140 data, FY2026 Q3, posted 2026-09-08; 28,874 EB-1A petitions were pending. Every quarter predates PA-2026-05 and Q4 was not posted as of 2026-11-01, so no data shows what the new policy has done. These are averages across all fields, not your odds.
Worked timeline: two ML researchers in November 2026
Two hypothetical researchers each file an EB-1A I-140 that USCIS receives on 2026-11-16, which becomes their priority date. Their queue generally follows country of birth, not citizenship.
| Step | Researcher A, born in India, H-1B at an industry lab | Researcher B, born in Brazil, STEM OPT after an ML PhD |
|---|---|---|
| November EB-1 Dates for Filing (the chart USCIS requires for I-485s) | 2024-07-01, so no I-485 yet | Current, so the chart allows a concurrent I-485 |
| November EB-1 Final Action Date | 2023-02-01 | Current |
| I-140 processing (USCIS tool, 2026-11-01, 80% of cases) | 32.5 months, or 15 business days with $2,965 premium | Same |
| Status bridge | An approved I-140 can support three-year H-1B extensions past six years under AC21 s.104(c) while the date is not current | F-1 requires a foreign residence you do not intend to abandon, so an immigrant petition can matter at a later visa interview or entry |
Researcher A's wait depends on the Visa Bulletin; the December bulletin was not out as of 2026-11-01, so any December figure is a forecast. Researcher B needs a licensed immigration attorney before filing an immigrant petition or I-485, and before any trip abroad while an I-485 is pending: for an F-1 student, leaving without advance parole generally abandons the I-485, and advance parole took 26.5 months in the USCIS tool on 2026-11-01. F-1 travel signatures and SEVIS questions go to the DSO. USCIS memo PM-602-0199 (in effect, challenged) also calls adjustment "a matter of discretion and administrative grace." Both should recheck the USCIS filing-charts page monthly.
O-1A: the nonimmigrant parallel for AI researchers
O-1A uses eight criteria in 8 CFR 214.2(o)(3)(iii)(B), the EB-1A list minus the two arts criteria, with "critical or essential capacity" in place of a leading or critical role. You need a major internationally recognized award or three criteria, then a totality review (Policy Manual, Volume 2, Part M, Chapter 4). Unlike EB-1A, it needs a US employer, a US agent, or a foreign employer through a US agent as petitioner, plus a written advisory opinion from a peer group unless none exists (USCIS O-1 page).
O-1 sits outside the H-1B lottery and outside Proclamation 11069, which covers H-1B only. USCIS approved 91.0% of O petitions (O-1 and O-2, initial and extensions) in FY2026 Q1-Q3 (USCIS I-129 data), a different form and measure, so it does not predict EB-1A. For O-1 extensions, DHS has proposed removing deference to prior approvals (RIN 1615-AD26); comments close 2026-11-13, current deference applies, and nothing changes unless a final rule is published. See O-1A vs EB-1A for sequencing.
Should you file EB-1A and NIW together?
Some researchers keep both pending. They are separate I-140s, each with its own fee ($715 paper or $665 online, plus an Asylum Program Fee of $600, $300 or $0 depending on the filer, per G-1055 edition 10/26/26), each decided on its own record. For India-born applicants, EB-1's Final Action Date (2023-02-01) is far ahead of EB-2's (2013-11-01); for the rest of the world, EB-1 is Current and EB-2 sits at 2025-04-01. Premium processing is 15 business days for EB-1A and 45 for NIW (USCIS premium processing). The NIW argument differs, too: see our EB-2 NIW plan for AI and machine learning engineers. One or both is a call for a licensed immigration attorney.
How an ML researcher's EB-1A record is usually assembled
- Export your publication and citation record from Google Scholar or Semantic Scholar, separating independent citations from self- and co-author ones.
- Download completed-review records from OpenReview, CMT or each journal's editorial system.
- Map each item to a criterion in 8 CFR 204.5(h)(3) using the Policy Manual chapter above.
- Document outside adoption of your work, such as GitHub dependents, product announcements and papers that build on it, with dates.
- Benchmark pay against position-appropriate compensation surveys for your role and location, not broad multi-occupation figures, with W-2s and offer letters.
- Request detailed letters for the I-140 exhibits from independent experts who know your work through its influence.
- Check chargeability and the current chart on the USCIS filing-charts page.
- Submit Form I-140 (edition 06/07/24) online, or by mail if you include Form I-907, as the USCIS I-140 page requires, with all required initial evidence.
This list describes the work; whether your record qualifies is a question for a licensed immigration attorney.
Common mistakes
- Counting criteria instead of arguing final merits. Three criteria reach step two; they do not win it.
- Leaning on coauthor and advisor letters. The officer weighs the witness's relationship to you.
- Citing a raw citation total without subfield benchmarks or self-citations removed.
- Filing reviewer invitations without completed reviews.
- Treating employer press releases as media about you.
- Leaving gaps in initial evidence. Under PA-2026-05 an officer can deny without asking for more.
- Comparing salary with the wrong group. The comparison is with others in your field.
- Claiming comparable evidence for a criterion that does apply, because the evidence for it is thin.
Frequently asked questions
Can an AI researcher get EB-1A without a major award?
Yes. Instead of a one-time major award such as a Nobel Prize, most ML researchers meet at least three of the ten criteria in 8 CFR 204.5(h)(3), usually authorship and judging plus original contributions, a critical role or a high salary. That only reaches final merits, which USCIS still applies as of November 2026.
Do NeurIPS or ICML papers satisfy EB-1A?
Peer-reviewed NeurIPS, ICML or ICLR papers usually satisfy authorship, met regardless of caliber. At final merits, influence, subfield-relative citations and independent adoption matter more than paper count.
Does reviewing for conferences count as judging?
It can, if you show you were invited and actually reviewed, so keep completed-review records. At final merits, area chair or editorial board roles generally carry more weight than a standard reviewer slot.
How many citations do I need for EB-1A?
There is no number. The Policy Manual asks for citations at a level indicative of major significance and, at final merits, a citation rate high relative to others in the field, with the methodology explained. Benchmark against your subfield and separate out self- and co-author citations.
Should I file EB-1A and NIW together?
Some researchers keep both pending as separate I-140s with separate fees, each decided on its own record. In April to June 2026, EB-1A approval was 32.2% and NIW 55.3%, both before PA-2026-05, and for India-born applicants EB-1 Final Action Dates are far ahead of EB-2 in the November 2026 bulletin. Decide with a licensed immigration attorney.
Sources
- USCIS Policy Manual, Vol. 6, Part F, Ch. 2 and Vol. 2, Part M, Ch. 4
- USCIS EB-1 page and O-1 page
- USCIS I-140 data, FY2026 Q3 and I-129 data, FY2026 Q3
- Analysis of Agarwal and Mukherji (Cyrus D. Mehta & Partners blog)
- PA-2026-05 and PM-602-0199
- Visa Bulletin and USCIS filing charts
- Form I-140, G-1055, premium processing, processing times
- Deference NPRM, 2026-10-14
How we checked this: on 2026-11-01 we read the USCIS Policy Manual chapters and every quotation against them, the EB-1, O-1, I-140 and premium processing pages, G-1055 (edition 10/26/26), the November 2026 Visa Bulletin values, the USCIS filing-charts page and the processing times tool, and recomputed each approval rate from the USCIS I-140 and I-129 spreadsheets.
This is general information, not legal or tax advice; your own notices and I-94, and your attorney (or DSO, for anything F-1, OPT or SEVIS), govern.
If you are on OPT, STEM OPT or H-1B and want a research role with an employer that sponsors, F1Jobs can help you run a focused search while you build your record.
Frequently asked questions
Can an AI researcher get EB-1A without a major award?
Yes. The regulation offers two routes, a one-time major internationally recognized award such as a Nobel Prize, or evidence meeting at least three of the ten criteria in 8 CFR 204.5(h)(3). ML researchers typically use the second route, usually authorship and judging plus original contributions, a critical role or a high salary. Meeting three criteria only brings you to the final merits review, which USCIS still applies as of November 2026.
Do NeurIPS or ICML papers satisfy EB-1A?
Peer-reviewed papers at venues such as NeurIPS, ICML or ICLR usually satisfy the authorship criterion, which the USCIS Policy Manual says is met regardless of the caliber of the publication. At the final merits step the officer asks whether those papers show you are at the very top of the field, so influence, citations relative to your subfield and independent adoption matter more than paper count.
Does reviewing for conferences count as judging?
It can. The USCIS Policy Manual says you must show you were invited to judge and actually did the judging, so keep completed-review confirmations, not only invitations. At final merits, roles that are harder to get, such as area chair, senior program committee or journal editorial board, generally carry more weight than a standard reviewer slot.
How many citations do I need for EB-1A?
There is no number. Neither 8 CFR 204.5(h) nor the USCIS Policy Manual sets a citation threshold. For original contributions, the Policy Manual looks for work cited at a level indicative of major significance in the field; at final merits, a citation rate that is high relative to others in the field, such as a high h-index, is a positive factor, and a petitioner relying on citation tools should explain the methodology. Benchmark against your own subfield and separate out self-citations and co-author citations.
Should I file EB-1A and NIW together?
Some researchers keep an EB-1A and an EB-2 NIW petition pending at the same time; they are separate I-140s with separate fees, each decided on its own record. In USCIS data for April to June 2026, EB-1A approval was 32.2% and NIW 55.3%, both before PA-2026-05, while EB-1 Final Action Dates are far ahead of EB-2 for India-born applicants in the November 2026 Visa Bulletin. Whether to file one or both is a decision to make with a licensed immigration attorney.