AI Safety & Alignment Researcher
Work out why models do what they do, and make them do it safely: interpretability, oversight, evaluations and red-teaming for frontier systems.
Code WORKWITHUS unlocks the O1DMatch Starter plan at no cost. The eligibility check is an AI review of your background, not a legal opinion.
| Field | AI Research |
|---|---|
| Usual visa path | O-1A (extraordinary ability) |
| Who hires this profile | Frontier labs, safety research nonprofits, government AI institutes' contractors |
| Market pay (data, not an offer) | USD 280k–750k+ total compensation |
| Work type | Full-time roles, contracts and short gigs with U.S. companies |
| How to get matched | Free eligibility check, then a profile with code WORKWITHUS |
About the role
This profile is written for people already doing this work at a high level. U.S. employers that hire it include frontier labs, safety research nonprofits, government AI institutes' contractors. The sections below cover the work, what those employers screen for, and how a career like this maps to the O-1.
What you would do
- Build interpretability tools (probing, sparse autoencoders, activation steering) and use them on real models
- Design evaluations for dangerous capabilities, deception and robustness
- Write the research that shapes how labs train and deploy models
- Partner with policy and product teams on what the results mean
What U.S. employers look for
- A strong research record in ML, even if your title was not 'safety'
- Deep ML fundamentals and comfort with large codebases
- Published or widely read safety work (papers, technical posts, released tools)
- Clear writing; this field moves on memos as much as papers
How this career maps to the O-1
An O-1 is judged against your field. People in this role usually build their case from evidence like this:
Pay
USD 280k–750k+ total compensation. Source: public H-1B LCA disclosures and levels.fyi, compiled May 2026; base + bonus + equity, approximate. This is market data to help you benchmark, not an offer from O1DMatch or any employer.
Visa path
O-1A. Safety research produces unusually public evidence, which makes the criteria easier to document.
Fit this profile? Here is how to get matched
- Run the free eligibility checkA short AI review of your background against the eight O-1 criteria. Most of what you enter carries into your profile.
- Create your profile with code WORKWITHUSTell us your field and the work you do today; every O-1 case is judged against your field.
- Upload your evidenceAward letters, press about you, contracts, pay records, judging invitations. Each file is sorted under the criterion it supports, and your criteria map shows what you already meet and what is one document away.
- Meet U.S. companiesEmployers on the platform post full-time roles and short gigs. Gigs you confirm with verified companies count toward the U.S. itinerary your petition needs.
- Get to an attorney-ready caseIf you have no U.S. employer to file for you, the platform shows agents that can petition. When your score reaches 75 with the key pieces in place, we connect you with an immigration attorney.
Code WORKWITHUS unlocks the O1DMatch Starter plan at no cost. The eligibility check is an AI review of your background, not a legal opinion.
Listings in this role family
Each listing is modeled on a real U.S. posting; the employer is not named. Apply through O1DMatch to be matched with employers hiring for this kind of role.
More AI Research role profiles
Foundation Model Research Scientist
Train and study the large language, vision and multimodal models that frontier labs ship. You own a research agenda, a…
O-1AFounding AI Engineer / Head of AI
Be the first dedicated AI hire at a venture-backed company: pick the stack, ship the first model-powered product, and…
O-1AAI Red Team / ML Security Researcher
Break language models before attackers do: prompt injection, jailbreaks, model extraction, data poisoning, and the…
How the agent-based O-1 works, from law firms
You do not need a single employer to file for you. A U.S. agent can petition for work with several employers. These explainers come from law firms, including ours, and from the regulation itself:
Read our plain-language guide →
Common questions
Is this a job opening?
No. This is a role profile: a description of a kind of role U.S. employers hire international talent for. It is not a vacancy at a specific company. If you fit it, O1DMatch helps you build the O-1 case and meet U.S. companies that post real roles and gigs on the platform.
Do I need one employer to file my petition?
Not necessarily. An O-1 must be filed by someone in the U.S., but that can be a U.S. agent filing for work with several employers. If you have no employer who can file, the platform shows agents that can petition on your behalf.
What does code WORKWITHUS give me?
It unlocks the O1DMatch Starter plan at no cost: your full eligibility breakdown and recommendations, your evidence map, and access to apply to roles and gigs from U.S. companies on the platform.
Does a high score mean my visa will be approved?
No. The score shows how your evidence lines up against the criteria. USCIS decides every case, and an attorney reviews yours before anything is filed. No outcome is promised.