The first thing an AI engineer usually says in an EB-1A consultation is some version of, "I think I might qualify, but I'm not sure." Then they list off a fairly familiar set of accomplishments. A few papers. Some open-source projects have respectable traction. A senior title at a company you've probably heard of. Maybe a patent or two. And then the real question, usually phrased as, "but am I, like, actually good enough for this?"
That conversation has gotten a lot more common in the past 18 months. Not because the rules changed, they haven't, but because the AI hiring boom collided with H-1B uncertainty and a green card backlog that, for some nationalities, is now measured in decades. EB-1A is starting to look like the exit ramp. For some people, it genuinely is. For others, it's a fast way to spend a lot of money on a petition that won't hold up.
This post is for figuring out which one you might be.
Why everyone in AI is suddenly thinking about EB-1A
Three things, mostly.
You don't need an employer to sponsor you. No PERM, no labor certification, no job offer. You file for yourself.
Priority dates are currently current for most countries, though it's worth checking the latest Visa Bulletin for your specific situation.
Premium processing is available on the I-140, so USCIS will give you a decision in 15 business days.
For an Indian-born ML engineer staring at an EB-2 wait that may outlast their career, that combination is hard to ignore. The catch is that EB-1A is also one of the most heavily scrutinized employment-based categories USCIS adjudicates. The bar is real, and plenty of strong-on-paper profiles still come back with a Request for Evidence or a denial.
How USCIS actually decides
There are two parts to the analysis, and most online guides skip the second one.
The first part is the criteria check. You either have a one-time major international award like a Nobel or Turing, which is rare, or you meet at least three out of ten regulatory criteria. Original contributions of major significance, authorship of scholarly articles, serving as a judge of others' work, leading or critical role at a distinguished organization, high salary, and a few more.
The second part is the "final merits determination," and this is where a lot of petitions quietly fall apart. Even if you've technically checked three boxes, USCIS then steps back and asks whether, taken together, your evidence actually shows you belong to that "small percentage at the very top of the field." Three weak criteria don't add up to one strong case. The petition has to tell a story about impact, not just compile a list of achievements.
What tends to work for AI profiles
Out of the ten criteria, five tend to carry the weight in AI cases.
Original contributions of major significance is usually the centerpiece. Novel methods that show up in downstream research, open-source tools with real adoption (not just GitHub stars, actual usage), patents that someone outside your employer has built on, architectures or techniques that have measurably changed how others in the field work.
Scholarly authorship matters when the venues do. NeurIPS, ICML, ICLR, ACL, CVPR carry weight. So do well-cited preprints with a clear citation trajectory. Workshop papers and posters help, but they don't anchor a case.
Peer review comes up more than people expect. Reviewing for top conferences and journals, or sitting on technical review committees. What USCIS is really evaluating is whether the field chose you because it considers you qualified to evaluate others.
Leading or critical role is where a lot of senior engineers think they're stronger than they actually are on paper. Your title doesn't do the work. You need contemporaneous documentation showing what you owned, what depended on you, and why the organization itself has a distinguished reputation.
High salary is easy to assume is automatic for AI engineers. It isn't. The comparison has to be against the right peer group using properly sourced wage data, not your offer letter.
We sometimes build around press coverage, association memberships, or awards. Even if they satisfy regulatory criteria, those rarely hold up an AI petition on their own.
What's going wrong in the RFEs we see
A few patterns, over and over.
Petitions that assert significance instead of proving it. "Widely adopted" with no third-party evidence doesn't pass.
Citation metrics presented without context. An h-index that's impressive in one subfield can be unremarkable in another, and USCIS won't do that interpretive work for you.
Recommendation letters that read like LinkedIn endorsements. Letters from former managers get discounted. What carries weight is independent experts, meaning people you haven't worked with, speaking to specific contributions and their effect on the field.
And the most common one: confusing the prestige of where you work with evidence of your own extraordinary ability. Working at a well-known AI lab is not, by itself, an EB-1A criterion.
So, are you EB-1A material?
A few honest screening questions before anyone spends real money on a petition.
Has your work been used or built on by people outside your own team and company? Can you point to independent experts who would credibly say you're at the top of your field? Is there documented, external evidence of your impact, or only your own description of it?
If most of the answers are "yes, but I'd need to gather the proof," you're probably worth a real conversation. If most of them are "not really, but I'm working on it," the more useful question is timing. EB-1A rewards patience. A case filed a year too early is much harder to recover from than the same case filed a year later with a stronger record.
A weak petition isn't just a denial. It's a paper trail that follows you. The petitions that succeed are usually the ones where strategy came first, and evidence-gathering came second.
