AI Safety Careers

Warm-Intro Onto an AI Safety Team at Anthropic or OpenAI

Cold applications to safety roles get buried under thousands of resumes. The honest path is a two-line note to someone who already shares a paper, a lab, or a reading group with you.

Find your warm path in →

Free · No sign-up · See results in ~10 seconds

A plain LinkedIn search buries the people who would actually help you.

It ranks by prominence, so the most-followed, most-messaged names sit on top. FindWarmIntros surfaces the coworkers who share your school or a past employer - the people most likely to reply - and hands you a ready-to-send intro.

Example - what you’ll see
in
Someone who works at your target company
🎓 Same university as you  ·  💼 Shared past employer
🔥 Strongest
in
A recruiter at your target company
🎓 Same university as you
🎓 Alumni

✍️ Ready-to-send intro“Hi - we both studied at [your school]. I’d love to hear about your path to a company you are targeting before I apply…”

… plus everyone else in your network who can put in a good word.

See who can refer you in - pick your target company:

Short answer: Map the specific team you want (interpretability, scalable oversight, RL from human feedback) and find its actual researchers on arXiv and the Alignment Forum. Then trace one warm path: a co-author you share, a former labmate who now works there, or someone who left the lab and will talk freely. Send them a 30-second note that leads with the thing you share and a single concrete question, not your resume.

Why cold safety applications almost never land

AI safety hiring at Anthropic and OpenAI is bottlenecked on trust and taste, not headcount. A recruiter cannot tell from a resume whether you actually understand why reward hacking is hard or whether you just read the abstract. A researcher on the team can tell in one message, and that researcher is the person whose opinion moves your application.

So the goal is not to apply harder. It is to get one person inside the safety org to read three sentences you wrote and think, this person gets it. That happens through people who can already vouch for the shape of your thinking: co-authors, labmates, and alumni.

Three warm paths onto an alignment team

Pick the path where you already share something real. Do not manufacture a connection you do not have, safety researchers notice.

  • Shared co-authors and acknowledgments. Open the papers from the exact team (Anthropic interpretability, OpenAI superalignment successors) and read the author list and the acknowledgments footnote. If you have ever co-authored with, been advised by, or thanked one of those people, you have a direct bridge. Ask that shared person for a one-line intro, not a job.
  • Lab and fellowship alumni. MATS, ARENA, Redwood, the Anthropic Fellows program, and academic labs like Berkeley CHAI feed both companies. Someone two cohorts ahead of you is now inside. Alumni help alumni because the shared program is a real credential, so lead with the cohort you have in common.
  • People who recently left the lab. Ex-researchers will tell you which team is actually hiring, what the interview really tests, and who to talk to, things no careers page says. They have no incentive to gatekeep and often know exactly which current researcher would find your work interesting.

Send a note that costs 30 seconds to answer

The mistake is asking a busy researcher to evaluate your whole candidacy. Instead, name the thing you share in the first line (the paper, the cohort, the reading group), then ask one specific question they can answer in a sentence: is the interpretability team hiring for the residency track this cycle, or would you be open to a five-minute question about the oversight team.

This is the part FindWarmIntros automates. You give it a target like Anthropic or OpenAI and it searches for the real people there who overlap with your school, past labs, or employers, then ranks them by how warm the path actually is and drafts the opening line that names what you share. It does the finding-and-ranking so you can spend your time on the one message that matters. If you would rather do it by hand, the same three paths above still work, it just takes longer.

Once someone replies, do not pivot straight to your resume. Ask what problem the team is stuck on and whether your recent project (a real one, linked) is relevant. A referral you earned by being useful beats one you extracted by asking.

Frequently Asked Questions

Should I message the team lead or a junior researcher?
Start with whoever you share the most concrete thing with, usually a mid-level researcher or a recent alum, not the team lead. A peer-level co-author who can say I know their work carries more weight than a cold note to a director who gets dozens a day.
What if I have never published a safety paper?
You do not need one. Shared context can be a fellowship cohort, a reading group, an open-source contribution to a safety eval, or a labmate who moved to the company. Lead with genuine shared work in adjacent ML, then show you have engaged seriously with the safety literature.
Is it worth reaching out to researchers who left the lab?
Yes, often more than current staff. Ex-researchers answer honestly about which team is hiring and what the interview tests, and many will point you to a specific current researcher who would find your work interesting. They have no reason to gatekeep.
How do I find who is actually on the safety team?
Read the recent papers from that specific team on arXiv and the author lists, follow the Alignment Forum and the labs published research index, and check fellowship alumni pages. FindWarmIntros can shortcut this by surfacing the people there who overlap with your background.

Keep going

Find your warm path in →
Find your warm intro →