O-1A Guide
How to Build an O-1A Petition for a Researcher in AI Safety and Alignment Whose Contributions Are Primarily in Technical Reports and Preprints
AI safety and alignment researchers often publish primarily through preprints and technical reports rather than peer-reviewed journals. This guide explains how to build an O-1A petition around that publication record, document original contributions through citation evidence, and satisfy the scholarly articles criterion despite non-traditional venues.
Why publication format matters for O-1A purposes
AI safety and alignment research is a field in which the dominant publication culture differs substantially from traditional academic disciplines. Major organizations active in the field — Anthropic, OpenAI, DeepMind, the Machine Intelligence Research Institute, the Center for Human-Compatible AI at UC Berkeley, and the Center for AI Safety — release much of their research as technical reports, preprints, and blog posts rather than exclusively in peer-reviewed conference proceedings or journals. The regulatory standard for the scholarly articles criterion under 8 C.F.R. § 214.2(o)(3)(ii)(A)(6) requires publications in professional journals or other major media in the field. The question an O-1A petition must answer is whether technical reports and preprints qualify as major media under this standard — and the answer depends on the specific publication channel, its institutional affiliation, and how the field treats the contribution.
USCIS and the AAO have not definitively ruled that arXiv preprints constitute major media under the scholarly articles criterion, but they have accepted the argument in cases where the petition established that arXiv is the primary rapid dissemination channel for the field, that papers posted on arXiv are cited and relied upon by other researchers before formal publication, and that the specific preprints at issue have achieved citation records and field influence comparable to peer-reviewed publications. For AI safety and alignment, this argument is well-supported: arXiv is the standard first-release channel for alignment research, and papers posted there are regularly cited, implemented, and discussed in the technical community before, and sometimes instead of, journal publication.
The petition should not rely solely on arXiv preprints if the petitioner has any peer-reviewed publications. Papers published in the proceedings of NeurIPS, ICML, ICLR, ACL, EMNLP, or the AAAI Conference satisfy the scholarly articles criterion without ambiguity, because these venues employ selective peer review and are recognized as the major publication venues in machine learning and AI research. The petition strategy should lead with peer-reviewed publications where they exist, treat technical reports from recognized organizations as supplementary scholarly articles evidence, and reserve the arXiv-as-major-media argument for petitioners whose record is primarily preprint-based.
Original contributions in AI safety research
The original contributions criterion under 8 C.F.R. § 214.2(o)(3)(ii)(A)(5) requires evidence that the petitioner's contributions have been of major significance. For an AI safety researcher, this criterion is among the strongest because the field is small enough that significant contributions are directly traceable and their influence is observable. A petitioner whose alignment technique — reward modeling, constitutional AI, scalable oversight, or interpretability methods — has been implemented or cited by other research teams, whose safety evaluation protocols have been adopted by organizations beyond the petitioner's own employer, or whose theoretical arguments about misalignment risks have been referenced in regulatory submissions, policy white papers, or industry safety frameworks has a demonstrable record of major significance.
Citation analysis from Google Scholar or Semantic Scholar is particularly informative for AI safety preprints because the field's citation practices are rapid and transparent. A technical report released by Anthropic, OpenAI, or a recognized academic group that accumulates hundreds of citations within a year of release — including citations in peer-reviewed conference papers, other organizational technical reports, and policy documents — has achieved a citation profile that demonstrates major significance regardless of formal journal publication. The petition should document the citation count, identify a representative sample of citing papers and their authors' institutional affiliations, and include an expert letter from a recognized researcher in the field who can explain why the cited work is considered significant.
Adoption of the petitioner's work in downstream systems or products is a form of original contributions evidence that is particularly relevant for AI safety researchers at organizations that deploy large-scale AI systems. A researcher whose safety evaluation methods are incorporated into the organization's model evaluation pipeline, whose alignment techniques are part of the organization's published model safety documentation, or whose interpretability tools have been released as open-source software with documentable usage metrics has contributions whose significance is observable at a concrete, non-speculative level. The petition should document these adoptions through organizational statements, published technical documentation, or GitHub repository metrics where applicable.
Scholarly articles and technical reports as major media
Technical reports from recognized AI research organizations occupy an intermediate position between peer-reviewed journal articles and informal blog posts. A technical report published by Anthropic, OpenAI, Google DeepMind, or the Center for AI Safety is produced by an organization with a distinguished reputation in the AI research community, is identified with a specific set of authors, is cited in subsequent research, and is publicly accessible through an institutional archive. A petition arguing that these reports constitute major media within the meaning of 8 C.F.R. § 214.2(o)(3)(ii)(A)(6) should establish the institutional reputation of the publishing organization, the circulation and citation record of the specific reports, and expert testimony that researchers in the field treat these reports as primary literature.
Peer-reviewed publications at NeurIPS, ICML, ICLR, ACL, the Workshop on Trustworthy and Socially Responsible Machine Learning (TrustML), the Workshop on Alignment for LLMs (ALM), or the FAccT (Fairness, Accountability, and Transparency) conference satisfy the scholarly articles criterion through the peer review process and the recognized standing of the venues. Acceptance rates at major AI conferences have remained selective — NeurIPS and ICML acceptance rates have hovered between 20 and 26 percent for the main track in recent years — making publication there a documented form of expert selection. The petition should document acceptance rates and the peer review process for each venue where the petitioner has published.
Publications in academic journals that cover AI safety or AI ethics — AI & Society, Ethics and Information Technology, the Journal of Artificial Intelligence Research, or Minds and Machines — also satisfy the scholarly articles criterion and may provide a useful complement to conference publications for petitioners whose work spans technical and policy dimensions of AI safety. A petitioner whose record includes both technical conference papers and journal publications addressing the societal dimensions of alignment research has a more complete scholarly articles argument than one whose record is confined to a single venue type. The petition should present the full publication record in a way that contextualizes the diversity of venues as reflecting the breadth of the field.
Judging and peer recognition structures
Peer review for AI safety and alignment research is structured differently from traditional academic fields. The major conference venues — NeurIPS, ICML, ICLR — use area chair and reviewer structures that involve thousands of reviewers; an invitation to review is common and does not on its own constitute strong judging criterion evidence. What distinguishes stronger judging evidence is selection for more senior or selective roles: area chair or senior area chair designation at a major AI conference, invitation to review for a grant program at NSF's Division of Information and Intelligent Systems or DARPA's Explainability, Fairness, and Accountability in AI programs, or invitation to serve on the program committee of a specialized AI safety workshop at a major venue that involves selectivity in reviewer recruitment.
The AI safety community has developed its own institutions for peer recognition and evaluation. Programs such as the Long-Term Future Fund, the Open Philanthropy AI Fellowship, and the Future of Life Institute's PhD Fellowship for AI safety researchers are explicitly competitive and recognition-based, making selection a form of peer recognition evidence. Similarly, invitation to present at the AI Safety Summit organized at Bletchley Park, the Machine Intelligence Research Institute's workshops, or the Center for Human-Compatible AI's annual retreat reflects a determination by the organizing institution that the petitioner's work is among the contributions the field's recognized experts consider worth presenting. These invitations and program committee roles satisfy the judging criterion through their structural function as peer evaluation activities.
Expert letters from recognized figures in AI safety and alignment — including faculty at recognized academic programs (UC Berkeley's Center for Human-Compatible AI, Oxford's Future of Humanity Institute, MIT's Computer Science and Artificial Intelligence Laboratory, Stanford's Human-Centered AI group) and senior researchers at recognized organizations — provide the interpretive framework for the judging and peer recognition evidence. A letter from a recognized researcher who has engaged with the petitioner's work — citing it, implementing it, or publicly discussing it — and who can attest to the petitioner's standing within the AI safety research community carries more weight than a letter from a senior colleague who simply endorses the petitioner's general competence.
Critical role and high salary evidence
The critical role criterion for an AI safety researcher at a recognized organization is developed through the petitioner's documented function within a specific research team or organizational unit. A researcher designated as the technical lead for a safety evaluation project at a major AI laboratory, the principal author of a model card or safety evaluation protocol that the organization has published as a commitment to external stakeholders, or the head of a specific research thrust such as interpretability, scalable oversight, or robustness evaluation occupies a critical role in an organization whose distinguished reputation is easily established through public documentation of its scale, funding, and research output. The employer's letter identifying the petitioner's specific function and explaining why it is non-interchangeable is the central piece of evidence.
For AI safety researchers at academic institutions, the critical role argument is grounded in grant records and laboratory leadership. A principal investigator on an NSF Responsible Computing or AI Ethics grant, a co-PI on an Open Philanthropy-funded research project, or a faculty member who leads the only AI safety research group at a recognized research university occupies a critical role in an institution with a distinguished academic reputation. The petition should include the grant documentation and a letter from the department chair or dean characterizing the petitioner's role within the institution's AI research portfolio. Early-career researchers who are not yet principal investigators can satisfy the critical role criterion through a key personnel designation on someone else's grant, supported by the PI's letter explaining the petitioner's essential function.
High salary evidence is a significant criterion for AI safety researchers employed by industry AI laboratories, because these organizations compete aggressively for safety research talent at compensation levels that substantially exceed academic market rates. A researcher employed at a major AI laboratory at a total compensation package — including base salary, equity, and bonus — that exceeds the 90th percentile for computer and information research scientists (BLS OEWS SOC code 15-1221) satisfies the high salary criterion with documentation from the employer. The petition should present total annualized compensation, not just base salary, and compare it to OEWS data for the relevant geographic market. For petitioners at academic institutions, the comparison is to faculty salary data from the AAUP's annual report for the appropriate institutional Carnegie classification.
Assembling the evidence file
An O-1A petition for an AI safety researcher whose primary outputs are technical reports and preprints should be designed with a clear awareness of the evidentiary risk profile. The scholarly articles criterion requires either peer-reviewed publications in recognized venues — which the petition should document first — or a sustained argument that the petitioner's preprints and technical reports constitute major media in the field. That argument must be made explicitly, supported by expert letters that attest to the field's publication norms, citation data demonstrating the circulation and influence of the specific works, and documentation of the publishing organization's standing within the AI research community. A petition that presents arXiv preprints without this scaffolding invites a request for evidence.
The petition should aim to satisfy at least four of the eight criteria: original contributions (most AI safety researchers can satisfy this with citation evidence and expert letters), scholarly articles (leading with any peer-reviewed publications, supplemented by technical reports), judging (conference review roles, grant panels, or program committee service), and critical role or high salary depending on the employment setting. The expert letters — three to five from recognized figures in AI safety or adjacent AI research fields — are the primary vehicle for contextualizing the field's publication norms and the petitioner's standing within the community. Letters should be solicited from researchers who have concrete, specific engagement with the petitioner's work, not just general familiarity.
A practical consideration for AI safety researchers is that many work at organizations that may impose confidentiality constraints on some research outputs, particularly safety evaluations of unreleased models. The petition should work with publicly available evidence wherever possible — published technical reports, arXiv preprints, conference papers, and public safety commitments the organization has made — and avoid relying on confidential internal documentation that USCIS cannot independently verify. Where confidential work is essential to the critical role argument, the employer's letter can describe the general nature and significance of the work without disclosing proprietary details. Premium processing under 8 C.F.R. § 103.7 is available and advisable given the typically time-sensitive employment situations of researchers in this rapidly developing field.
What we typically gather for this kind of case
| Document | Where to source | Why it matters |
|---|---|---|
| Peer-reviewed publications | Web of Science / Scopus exports | Anchors original-contributions and authorship criteria |
| Citation analysis | Google Scholar profile + ESI top-1% data | Quantifies major significance in the field |
| Salary benchmark | BLS OEWS for SOC code + locality | Documents high-salary criterion at 90th-percentile or above |
| Critical-role letters | Direct supervisor + program director | Establishes role's importance, not just title |
What we see go wrong, again and again
- 01Treating extraordinary ability as a credentials checklist rather than a story of field-wide impact.
- 02Submitting bibliometric data (h-index, citation counts) without explaining what makes those numbers high relative to peers in the same sub-field.
- 03Relying on letters from collaborators or co-authors rather than independent experts who can speak to influence.
See if you qualify
Lando reviews your background against the O-1A visa criteria and tells you honestly where you stand. Free, no commitment.