O-1A for AI and ML Engineers: How USCIS's Comparable Evidence Guidance Changed STEM Cases

A January 2025 USCIS policy update lets AI and ML engineers use GitHub activity and preprints as O-1A evidence, but the bar for a strong filing is higher than most job postings suggest.

If you're an AI or ML engineer weighing O-1A against the H-1B lottery, you've probably run into the same wall everyone in a fast-moving technical field hits: the visa's 8 criteria were written with scholars, athletes, and film directors in mind, and your actual proof of impact lives in a GitHub repo, a preprint server, or a model card nobody at USCIS has heard of. That mismatch used to leave adjudicators guessing. As of a Policy Manual update effective January 8, 2025, it doesn't have to.

USCIS now gives officers explicit instructions on how to evaluate "comparable evidence" for O-1A beneficiaries in fields like AI and machine learning, where the traditional criteria don't map cleanly onto founders, researchers, and engineers. This piece walks through what the update actually says, how it changes what you should submit, and where the risk still sits — including a newer, reported change to how USCIS handles thin filings.

The O-1A framework hasn't changed. The evidence you can use to meet it has.

Nothing about the legal test moved. To qualify for O-1A extraordinary ability, you still need to show either a single major internationally recognized award (a Nobel Prize is the classic example, though nothing that narrow is required in practice) or evidence satisfying at least 3 of 8 regulatory criteria, followed by USCIS's two-step "final merits" review of your record as a whole. The 8 criteria include things like national or international awards, membership in associations that demand outstanding achievement, published material about you in professional publications, judging the work of others, original contributions of major significance, authorship of scholarly articles, employment in a critical or essential capacity for a distinguished organization, and high salary relative to others in the field.

The problem for AI/ML engineers was never the legal standard — it was fit. A 26-year-old engineer whose "original contribution of major significance" is a widely-used open-source inference library doesn't look like a tenured professor's CV, even when the underlying influence is comparable or greater. The regulation has always permitted "comparable evidence" when a listed criterion doesn't readily apply to your occupation, but until the January 2025 update, officers had little concrete direction on what comparable evidence should look like for STEM founders, entrepreneurs, and researchers whose careers don't generate law-firm-style paper trails.

What the January 8, 2025 update actually instructs officers to do

The Policy Alert dated January 8, 2025 clarifies USCIS's approach in the O-1A section of the Policy Manual. Two changes matter most for STEM cases:

1. Officers are told to weigh the full record, not each criterion in isolation. A borderline showing on one criterion can be strengthened by the overall pattern of evidence in the final merits determination, rather than being knocked out early because it doesn't perfectly satisfy a single prong.
2. Nontraditional evidence is explicitly recognized as potentially comparable. The update names peer-reviewed preprints, GitHub contributions carrying citation-equivalent impact, and notable open-source projects as the kind of proof that can stand in for criteria written around more traditional academic or artistic careers.

That's a meaningful shift in emphasis, not a lowered bar. "Citation-equivalent impact" is doing real work in that sentence — a repo with a handful of stars from your own team doesn't clear it. What the guidance rewards is a petition that can show your open-source project, preprint, or technical contribution moved the needle for people who don't know you personally: independent adoption, forks and downstream dependents, technical citations in other researchers' papers, or write-ups by people you didn't ask to write about you.

Old approach vs. new approach for a typical AI/ML case

Building a comparable evidence case for GitHub and open-source work

Raw metrics rarely carry a petition by themselves. What tends to work is pairing the metric with independent corroboration that a reasonable adjudicator, who is not an ML specialist, can follow.

1. Identify which of the 8 criteria your evidence is standing in for. Comparable evidence has to map to a specific criterion — you're not inventing a ninth category. A widely-adopted library usually maps to "original contributions of major significance"; a preprint with citations usually maps to "authorship of scholarly articles" or "published material about you."
2. Document independent adoption, not just your own usage numbers. Companies or research groups that adopted your tool without you asking them to, dependent packages, and unaffiliated technical write-ups all carry more weight than raw star counts.
3. Get expert letters that explain the technical significance in plain language. A letter from a recognized researcher or engineer at another organization, describing specifically why your contribution mattered and to whom, does more work than a stack of screenshots.
4. Show a pattern across multiple criteria, not one strong data point. Given the final-merits instruction to weigh the whole record, a moderate showing on 3-4 criteria tends to outperform one spectacular criterion and two thin ones.
5. Keep a paper trail as you build, not just at filing time. Star history, citation alerts, and conference acceptance emails are far more persuasive collected contemporaneously than reconstructed months later.

If you're also comparing O-1A against the extraordinary-ability green card track, the eligibility overlap and the differences in what each requires are covered in O-1A vs EB-1A: extraordinary ability differences — worth reading before you decide which petition to prioritize first, since evidence built for one often supports the other.

What hasn't changed, and what to watch

A few things the comparable evidence update does not do:

• It does not create a lower overall standard. USCIS's O-1 policy guidance on uscis.gov still requires the same 3-of-8 or single-major-award threshold and the same final merits review.
• It does not guarantee that a given GitHub project or preprint will be accepted — officers retain discretion, and weak corroboration is still weak corroboration regardless of the category it's filed under.
• It does not change filing fees, processing times, or the underlying I-129 petition process.

One development worth flagging with appropriate caution: multiple immigration law firms have reported that, following an August 2026 policy update, USCIS officers now have greater discretion to deny a petition outright — without first issuing a Request for Evidence — when the initial filing is missing key documentation. This is reported by practitioners rather than a confirmed Policy Manual change at the time of writing, so treat it as an emerging practice, not settled law, and confirm the current posture with a licensed immigration attorney before you file. If accurate, it raises the cost of an underdeveloped comparable-evidence packet, since there may be no second chance to fill a documentation gap after filing.

Separately, if you're weighing O-1A against a founder-track filing, O-1 visa for startup founders and tech entrepreneurs covers how the same comparable-evidence logic applies when your "employer" is your own company, which raises its own agent-petitioner questions.

Fees to plan around

Beyond attorney costs and USCIS filing fees, budget for the Visa Integrity Fee, a $250 charge that applies to O-1 visa issuance effective October 1, 2025 under the One Big Beautiful Bill Act (OBBBA). This is charged at the visa-issuance stage for consular processing, separate from the I-129 petition fees paid to USCIS, and you can confirm current visa fee categories on travel.state.gov. If you need a refresher on the O-1 process end to end, including the standard fee stack, see O-1 visa complete guide 2026.

Common mistakes

• Treating GitHub stars as self-evident proof. A number without context (who adopted it, why, and independent of you) rarely persuades an officer who isn't an engineer.
• Filing comparable evidence without naming which criterion it replaces. Petitions that just attach a pile of technical artifacts without connecting them to a specific regulatory criterion make the officer do work they may not do in your favor.
• Assuming the January 2025 update lowers the bar rather than clarifies the path. It's guidance on how to evaluate nontraditional evidence, not a relaxed standard — final merits review still applies to the whole record.
• Skipping expert letters because the technical work "speaks for itself." It doesn't, to a generalist adjudicator. A well-written letter from an independent expert is often the difference between a comparable-evidence argument that lands and one that doesn't.
• Ignoring the STEM-specific evidence types the guidance names. If your case leans on GitHub or preprints, cite the January 8, 2025 Policy Manual update by name in your petition letter so the officer can see you're working within recognized guidance, not asking them to improvise.
• Not budgeting for the Visa Integrity Fee or confirming current wait times before booking a consular interview. Check travel.state.gov's visa wait times page before you plan travel around your interview date.

Frequently asked questions

What is comparable evidence under the O-1A criteria? O-1A regulations list 8 fixed criteria such as awards, published material, and high salary, but the rule always allowed a petitioner to submit evidence that is comparable when a listed criterion does not readily apply to their occupation. USCIS's January 8, 2025 Policy Manual update gives officers explicit direction on how to evaluate that comparable evidence in STEM fields, rather than creating a new legal standard.

Can GitHub contributions count as O-1A evidence for a software engineer? Yes, as comparable evidence when a traditional criterion like published material or original contributions does not map cleanly onto your work. The January 2025 guidance names GitHub contributions with citation-equivalent impact, notable open-source projects, and peer-reviewed preprints as the kinds of nontraditional evidence officers should weigh, but it still has to demonstrate a level of recognition consistent with extraordinary ability.

Does the January 2025 USCIS policy update change who qualifies for O-1A? It does not change the underlying statute or the requirement to meet at least 3 of 8 criteria or show one major internationally recognized award. What changed is officer guidance on how to weigh nontraditional STEM evidence and the instruction to review the totality of the record in the final merits determination rather than criteria in isolation.

What is the Visa Integrity Fee and does it affect O-1A petitions? The Visa Integrity Fee is a $250 charge that applies to O-1 visa issuance effective October 1, 2025 under the One Big Beautiful Bill Act. It is a fee tied to visa issuance at the consulate, separate from the USCIS filing fees for the underlying I-129 petition, so budget for it alongside your consular appointment costs.

Can USCIS deny an O-1A petition without issuing an RFE? Multiple immigration firms have reported that, following an August 2026 policy update, officers now have greater discretion to deny outright when a filing lacks key documentation, instead of issuing a Request for Evidence first. This is still an emerging, reported practice rather than a confirmed rule change, so confirm the current posture with a licensed immigration attorney before you file.

Sources

• USCIS Policy Manual, Volume 2, Part M — O and P Nonimmigrants
• USCIS Policy Manual updates and Policy Alerts
• Travel.state.gov — visa fees
• Travel.state.gov — visa appointment wait times

None of this is legal advice, and no petition strategy here is a guarantee of approval. Whether comparable evidence fits your specific record is a judgment call for a licensed immigration attorney who can review your actual documentation — not something to self-diagnose from a blog post. If you're building your case timeline around a job search on OPT or H-1B while you weigh O-1A, F1Jobs can help you think through where an extraordinary-ability track fits alongside your broader visa strategy.