Immigration · February 14, 2025 · 6 min read
EB-1 petitions for AI researchers: what works in 2025
USCIS adjudication standards for EB-1A and EB-1B have shifted meaningfully over the past 18 months — particularly for AI and ML researchers. We look at what the current bar actually is and how to clear it.
Two years ago, an EB-1 petition for a senior AI researcher was a relatively predictable filing. The category had standards, but the standards were navigable for the kind of candidate who was building production systems at a top lab. That has changed. The bar is higher; the evidence required is more specific; and adjudicators are reading the field differently than they did in 2023.
What the category actually requires
The category requires a showing of “extraordinary ability,” which the regulations operationalize as either (a) a one-time achievement of major international acclaim, or (b) satisfying at least three of ten enumerated criteria (8 CFR §204.5(h)(3)).
The category, which requires an employer petition, has a lower bar in theory — international recognition rather than extraordinary ability — but in practice the evidentiary standards have converged. A petition that would have cleared EB-1B easily in 2022 may now draw an RFE that, on its face, looks like an EB-1A challenge.
Where the standard has actually shifted
Citation counts no longer carry the case
In 2022 and earlier, a petitioner with several thousand Google Scholar citations could point to that volume and expect it to substantiate the “original contributions of major significance” criterion. That is no longer reliable. Adjudicators are increasingly looking at the contextual significance of citations — specifically, whether the work has been adopted, extended, or cited in a way that demonstrates impact rather than ubiquity.
Industry contributions are getting more weight
A welcome shift: contributions in industry — published model releases, widely-adopted open-source frameworks, leadership in collaborative research — are now being read more generously than they were in 2022. The traditional evidentiary frame was academic publishing; petitions for industry researchers historically had to translate impact into academic terms. The translation requirement has softened, particularly where the researcher led or co-led an artifact (a model, a benchmark, a tool) that the field uses.
Letters of support are scrutinized harder
Recommendation letters from senior figures — once a near-default piece of evidence — are now scrutinized for specificity. Boilerplate language is read as boilerplate. The letters that move the needle now are detailed: they describe specific contributions, specific impact, and specific knowledge of the petitioner's work, written by figures whose own standing is independently established.
The 2024-2025 RFE patterns
The most common RFE we are seeing on AI EB-1 petitions hits the “original contributions of major significance” criterion. The challenge is almost always framed as: You have demonstrated that the petitioner has done significant work; you have not demonstrated that the work has had an impact that elevates it above the field as a whole.
The response strategy that works is concrete: cite specific instances where the petitioner's work has been adopted by others (other models built on it, papers extending it, products shipping with it), and where reasonable, document the downstream impact in dollar or user-count terms.
What we recommend candidates do
Document in real time
The biggest gap we see in AI petitions is that researchers do not document their own impact contemporaneously. By the time we are filing, we are reconstructing a record. A researcher who maintains a list of: (a) artifacts they led or contributed to materially, (b) downstream usage they can verify, and (c) substantive citations of their work has cut weeks of preparation off the petition timeline.
Plan the support letter slate early
Eight to ten letters from carefully chosen referees produce stronger petitions than fifteen letters that include generic supporters. Plan who is going to write, what each letter is going to say (so they don't overlap), and approach the strongest referees first.
Take peer review seriously
The “judging the work of others” criterion — peer review for top conferences, program committee membership, journal review — is one of the few criteria where the evidence is unambiguous and self-authenticating. Researchers who have not yet taken on peer review work should consider it both for the substantive contribution and for the evidentiary value.
The bottom line
EB-1 petitions for AI researchers are still very much approvable in 2025; the median processing time on the petitions we have filed in the past nine months has been seven months from filing. But the bar is real, and the evidence has to be specific. Petitions that worked on volume in 2022 do not work on volume now. They work on specificity.