EB-2 NIW for AI and Machine Learning Engineers: An Evidence Plan for Industry, Not Just Labs
USCIS policy names certain areas of AI as examples of critical and emerging technology, yet says gains to your employer alone do not count. Here is the evidence plan for the gap between the two.
Part of our guide to EB-1A and EB-2 NIW.
Short answer, as of 2026-11-01: yes. An EB-2 NIW for machine learning engineers can rest on industry work, without a PhD or a long publication list, but only if the petition proves a specific endeavor whose impact reaches past your employer. The USCIS Policy Manual (Volume 6, Part F, Chapter 5, current as of 2026-10-27) names "certain critical areas of artificial intelligence" among its examples of critical and emerging technology. The same chapter says benefits to one employer alone "are not sufficiently relevant." Your evidence plan lives in the space between those two sentences.
Status as of 2026-11-01:
- In effect: the Matter of Dhanasar three-prong test, as applied in the Policy Manual chapter on national interest waivers.
- In effect, challenged: PA-2026-05 (since 2026-08-05) lets officers deny any benefit request, NIWs included, without an RFE or NOID when required initial evidence is missing. AAUP v. Mullin (D. Mass., filed 2026-10-05) challenges it; no ruling or injunction yet.
- Latest data: NIW approval rate 55.3% in FY2026 Q3 (April to June 2026), up from 42.5% in Q1, all before PA-2026-05.
- Processing (USCIS tool, read 2026-11-01): NIW I-140, 30 months; premium processing, 45 business days for a $2,965 I-907 fee.
- Under OIRA review, unpublished: "Amending the Premium Processing Timeframe" (RIN 1615-AD25), received 2026-10-13. No text is public; current timeframes apply.
Can a machine learning engineer get an EB-2 NIW?
An ML engineer can get an EB-2 NIW by qualifying for EB-2 and proving the three Dhanasar prongs: the specific AI or ML endeavor has substantial merit and national importance, the engineer is well positioned to advance it, and waiving the job offer benefits the United States. Industry work counts: the Policy Manual says STEM endeavors "whether in academic or industry settings" can show national importance.
For most industry engineers the hard part is prong one: proving that the work matters to a field or the public, not only to the company paying for it. Our guide to national importance examples across fields covers the general standard; this post maps it onto ML roles, where most of the work is proprietary.
First, the EB-2 baseline
The NIW waives the job offer and labor certification, not EB-2 itself. The USCIS EB-2 page asks for a US advanced degree or foreign equivalent, or a bachelor's followed by at least five years of progressive experience in the specialty. Exceptional ability (at least three of six criteria) is the other route. If your degree is in an unrelated field, review the fit with an immigration attorney first.
What USCIS policy says about AI endeavors
Three passages in the Policy Manual chapter carry most of the weight for ML petitions.
AI can be a critical and emerging technology, but only "certain" areas. Footnote 93 says examples "may include, but are not limited to, certain critical areas of artificial intelligence or quantum information science." To identify such a field, officers "consider governmental, academic, and other authoritative and instructive sources" along with your evidence. So "AI is important" proves nothing; the petition has to show, with sources of that kind, why your particular area matters to US security or competitiveness. For prong two, an advanced degree, "particularly a Doctor of Philosophy," tied to such a field is "an especially positive factor," though "a degree in and of itself" is not enough.
Your employer's gain does not count by itself. The chapter says: "Benefits to a specific employer alone, even an employer with a national footprint, are not sufficiently relevant to the question of whether a person's endeavor has national importance." It even names the weakest engineering case: "a software engineer adapting their employer's code for various clients will have difficulty demonstrating the national importance of that endeavor, absent additional broader impacts supported by specific evidence."
USCIS names the bridges. For technology built for one company, a petitioner could show "widespread interest in adoption or licensing of the technology, a novel and important manufacturing or operational process, or how the technology stands to impact the development of similar technology by other companies." Those three routes are the backbone of an industry ML petition.
One more rule shapes the drafting: the endeavor "is more specific than the general occupation." "Machine learning engineer" is an occupation; "building low-latency fraud-detection models for real-time payment networks" is starting to be an endeavor.
Evidence map: three ML profiles against the Dhanasar prongs
These are illustrations, not assessments of any case.
Evidence map for three common ML profiles, as of 2026-11 (Policy Manual current as of 2026-10-27).
| Prong | Applied ML engineer at a product company | Research engineer with papers | ML infrastructure engineer |
|---|---|---|---|
| 1. Merit and national importance | A named problem the models address across a sector (for example fraud in instant payments or triage in medical imaging), with impact shown beyond the employer | A research direction (for example efficient training or model evaluation) carried into products, with implications for the field | Systems that make training or serving models cheaper, faster or more reliable for others, such as an open-source runtime or compiler |
| 2. Well positioned to advance it | Patents or published applications; deployment results confirmed by an employer letter; letters from customers or partners who use the work; independent expert letters | Citation record; adoption of your methods by other groups; peer review; invited talks | Documented use by other organizations; maintainer role; integrations into other projects; conference talks |
| 3. On balance, waive the job offer | Benefit to the US even if other workers are available; knowledge or skills beyond an occupation's minimum requirements that a labor certification cannot capture | The same factors, plus research outputs that others build on | The same factors, with impact spread across many downstream users |
| Typical gap | Everything is proprietary, and impact is described as company revenue | Papers prove the past, but the endeavor reads as "keep doing my job" | Star or download counts with no proof of who relies on the work |
The applied engineer has the hardest prong one and the thinnest public record. The research engineer has the easiest prong two, but prong one still asks about the future endeavor. The infrastructure engineer often has the cleanest proof of use by others, if it is documented.
How do I show national importance for an AI product?
Separate the endeavor from the product's revenue. Show that what you are building changes something for a field, a region or the public, with evidence from outside your company. USCIS's own example is a drug developer who may establish national importance by showing "the prospective public health benefits of the drug" rather than only projecting the employer's profits.
For an ML product, that usually means four kinds of material:
- A field-level problem statement backed by government or academic sources on, say, payment fraud, grid reliability or diagnostic error.
- Third-party interest in adoption or licensing, such as partner agreements or integration requests your employer allows you to describe.
- Influence on other builders: evidence that other companies' or researchers' systems changed because of your method, model or tool.
- A route for continuing the work in the US, whether at "a particular employer or some other way," as the Policy Manual puts it.
Proprietary work: proving impact you cannot publish
Most of an industry ML engineer's best work sits behind an NDA, and papers are optional: the Policy Manual says its evidence list "is not meant to be a checklist or to indicate that any one type of evidence is either required or sufficient." The list asks for evidence of "how the person's work is being used by others, such as, but not limited to," contracts, documents showing technology you invented and how others use it, and patents or licenses "with documentation showing why the patent or license is significant to the field."
- Patents and published applications. A patent number shows you invented something; licensing, adoption or later work that builds on it shows why it matters.
- Adoption confirmed by outsiders. Your statement that a model cut fraud losses is a claim; a customer's or partner's letter describing what changed after they adopted it is corroboration. See independent expert letters versus collaborator letters for why the distinction matters.
- Open-source use by others. Open source is not named in the list, but "such as, but not limited to" leaves room for it. What persuades is documented reliance: downstream projects that depend on your library, integrations by named companies or labs, letters from independent maintainers. Stars and downloads show attention, not reliance.
- Invited talks and technical writing. The list includes "published articles or media reports about the person's achievements" and evidence that the work "has influenced the field of endeavor."
Two guardrails: get written clearance from your employer before any internal document goes into a petition, and make sure every number is true and documented.
Should an ML engineer file NIW or EB-1A?
It depends on your record and country of birth; the choice belongs with a licensed immigration attorney.
NIW vs EB-1A for ML engineers, as of 2026-11-01 (Final Action Dates from the November 2026 Visa Bulletin).
| EB-2 NIW | EB-1A | |
|---|---|---|
| Legal test | Three Dhanasar prongs | A one-time major achievement or 3 of 10 criteria, then a final merits review |
| Approval rate, FY2026 Q3 (Apr-Jun 2026, pre-PA-2026-05) | 55.3% | 32.2% |
| USCIS processing (tool, read 2026-11-01) | 30 months | 32.5 months |
| Premium processing | 45 business days, $2,965 | 15 business days, $2,965 |
| Final Action Date, India | 2013-11-01 | 2023-02-01 |
| Final Action Date, China | 2021-10-01 | 2023-08-01 |
| Final Action Date, rest of world | 2025-04-01 | Current |
Rates come from the USCIS FY2026 Q3 I-140 data (approvals over approvals plus denials); our NIW and EB-1A approval rate breakdown walks through the file. On EB-1A's final merits step, two 2026 District of Nebraska decisions split (Mukherji v. Miller ordered an approval; Agarwal v. Miller upheld the step), with no appellate precedent, so USCIS still applies it. Our comparison of EB-1A and EB-2 NIW for engineers covers the criteria side by side.
A worked timeline: one ML engineer, three birth countries
Take a hypothetical ML engineer on H-1B with a US master's in computer science. Their self-petitioned NIW I-140 reaches USCIS on Monday, 2027-01-11, with a Form I-907. The USCIS I-140 page says an I-140 filed together with an I-907 must go by mail.
- 2027-01-11: because an NIW needs no labor certification, the I-140 filing date becomes the priority date.
- Around 2027-03-17: the 45th business day after receipt, skipping the 2027-01-18 and 2027-02-15 federal holidays, if the timeframe is unchanged by then. The USCIS premium processing page says USCIS will issue an approval, denial, NOID or RFE, or open a fraud investigation, in that window; an RFE or NOID stops the clock, and a new period starts when USCIS receives the response.
- Without premium: the tool showed 30 months on 2026-11-01. That figure moves; check it again before deciding.
After approval, birth country decides what comes next. In the November 2026 bulletin (Final Action / Dates for Filing), EB-2 India is 2013-11-01 / 2015-01-15, so an India-born engineer's approved I-140 mainly holds the priority date and, under AC21 section 104(c), can support three-year H-1B extensions beyond the six-year limit. EB-2 China is 2021-10-01 / 2023-01-01. Rest of world is 2025-04-01 / 2026-03-15, so a 2027 priority date waits until the chart USCIS designates reaches it (for November, Dates for Filing). Consular cases always use Final Action Dates. The December bulletin was not out as of 2026-11-01, so any later cutoff is a forecast. If you still need a sponsoring employer, our guide to machine learning engineer visa sponsorship covers the H-1B and O-1 routes.
Step by step: an EB-2 NIW evidence plan for machine learning engineers
- Confirm EB-2 eligibility against the USCIS EB-2 page: degree records, a credential evaluation for a foreign degree, and experience letters if you rely on a bachelor's plus five years.
- Draft the endeavor statement: the problem, the field, how you will advance it in the US, and why it reaches beyond one employer, citing government and academic sources on why your AI area matters.
- Collect outside proof of use: USPTO patent records, partner or licensing documents your employer releases, open-source dependents and citations.
- Request letters from independent experts and from users outside your employer; the Policy Manual says letters "should be supported by other independent evidence."
- Prepare Form I-140 (edition 06/07/24) with a completed Form ETA-9089 Appendix A and a signed ETA-9089 Final Determination, which the USCIS EB-2 page requires for NIW petitions. Fee: $715 paper or $665 online, plus an Asylum Program Fee of $600, $300 or $0 depending on the petitioner (G-1055 edition 10/26/26).
- Decide with your attorney whether to add Form I-907. Under PA-2026-05 a faster answer can also be a faster denial.
- Check the Visa Bulletin and the USCIS filing chart before any I-485; a self-petitioner can file it with the I-140 only when their priority date is current under the designated chart.
- Ask an immigration attorney about immigrant intent if you are on F-1 STEM OPT, and your DSO about anything touching SEVIS or travel. F-1 is not a dual-intent status; our guide to filing an I-140 while on OPT explains where the exposure sits.
Common mistakes
- Describing the job, not the endeavor. "I will keep working as an ML engineer at my company" names an occupation and an employer, not an endeavor.
- Treating "AI" as automatic national importance. The Policy Manual says "certain critical areas" of AI; show with authoritative sources why yours is one.
- Filing only collaborator letters. Letters from your manager and co-authors explain the work, but independent evidence has to back them.
- Filing thin and waiting for an RFE. Under PA-2026-05, missing required initial evidence can mean a denial with no RFE.
- Misconception: "NIW approvals collapsed after PA-2026-05." No USCIS data covers decisions under the policy; the latest quarter (April to June 2026) shows 55.3%.
Frequently asked questions
Can an AI engineer without publications get an NIW?
Publications are not required. The USCIS Policy Manual (current as of 2026-10-27) says its evidence list is not a checklist and that no single type of evidence is required. Patents, documented use of your work by other organizations and detailed independent letters can carry the second prong, but the endeavor must still show national importance beyond your employer.
Do open-source contributions count as NIW evidence?
They can, as evidence that others use your work, which the Policy Manual counts for the second prong. Documented use by other organizations, such as downstream projects, integrations or letters from independent maintainers, carries weight; star or download counts alone do not. Open source does not replace the first-prong case for national importance.
How do I show national importance for an AI product?
Show impact beyond the company that sells it. As of 2026-11-01 USCIS policy says benefits to a specific employer alone are not sufficiently relevant, and points to evidence such as widespread interest in adopting or licensing the technology, or its effect on how other companies build similar technology. Tie the endeavor to a field-level problem documented by government or academic sources.
Should an ML engineer file NIW or EB-1A?
It depends on your record and country of birth, so ask a licensed immigration attorney. In USCIS data for April to June 2026, before PA-2026-05, NIW approvals ran at 55.3% and EB-1A at 32.2%. For India-born filers the November 2026 Final Action Date is 2023-02-01 for EB-1 versus 2013-11-01 for EB-2, but EB-1A has a higher evidence bar.
What is the NIW approval rate in 2026?
USCIS quarterly I-140 data shows 42.5% in FY2026 Q1, 48.1% in Q2 and 55.3% in Q3 (April to June 2026), counting approvals over approvals plus denials. All of those decisions predate PA-2026-05, in effect since 2026-08-05, and no later quarter was posted as of 2026-11-01. A quarterly rate is not your personal odds.
Sources
- USCIS Policy Manual, Vol. 6, Part F, Ch. 5 (current as of 2026-10-27) and PA-2026-05
- USCIS EB-2, Form I-140, premium processing and G-1055
- USCIS I-140 data, FY2026 Q3 and processing times
- USCIS filing charts and the State Department Visa Bulletin
- OIRA, RIN 1615-AD25
How we checked this: we read the Policy Manual chapter, the USCIS EB-2, I-140 and premium processing pages, PA-2026-05, the FY2026 Q3 I-140 data, G-1055 and the processing-times tool on 2026-11-01; Visa Bulletin values are the November 2026 bulletin (released 2026-10-22), cross-checked against several law-firm summaries.
This is general information, not legal or tax advice. Your own notices govern, and an immigration attorney (or your DSO for anything F-1 or SEVIS) should review your specific facts.
If you are an ML engineer on STEM OPT or H-1B and want your job search to point toward employers that sponsor and value this kind of work, F1Jobs can help you plan it.
Frequently asked questions
Can an AI engineer without publications get an NIW?
Publications are not required. The USCIS Policy Manual (current as of 2026-10-27) says its evidence list is not a checklist and that no single type of evidence is required. Patents, documented use of your work by other organizations and detailed independent letters can carry the second prong, but the endeavor must still show national importance beyond your employer.
Do open-source contributions count as NIW evidence?
They can, as evidence that others use your work, which the Policy Manual counts for the second prong. Documented use by other organizations, such as downstream projects, integrations or letters from independent maintainers, carries weight; star or download counts alone do not. Open source does not replace the first-prong case for national importance.
How do I show national importance for an AI product?
Show impact beyond the company that sells it. As of 2026-11-01 USCIS policy says benefits to a specific employer alone are not sufficiently relevant, and points to evidence such as widespread interest in adopting or licensing the technology, or its effect on how other companies build similar technology. Tie the endeavor to a field-level problem documented by government or academic sources.
Should an ML engineer file NIW or EB-1A?
It depends on your record and country of birth, so ask a licensed immigration attorney. In USCIS data for April to June 2026, before PA-2026-05, NIW approvals ran at 55.3% and EB-1A at 32.2%. For India-born filers the November 2026 Final Action Date is 2023-02-01 for EB-1 versus 2013-11-01 for EB-2, but EB-1A has a higher evidence bar.
What is the NIW approval rate in 2026?
USCIS quarterly I-140 data shows 42.5% in FY2026 Q1, 48.1% in Q2 and 55.3% in Q3 (April to June 2026), counting approvals over approvals plus denials. All of those decisions predate PA-2026-05, in effect since 2026-08-05, and no later quarter was posted as of 2026-11-01. A quarterly rate is not your personal odds.