AIS Hiring Snapshot

The entry-level cliff in AI safety hiring

Pulled straight from the orgs' own hiring systems, today's open roles skew sharply senior — early-career entry points are scarce.

Snapshot 2026-06-17 · 69 open roles across 8 AI-safety organizations with public feeds · frontier labs shown separately, not blended in.

6%
of open roles are accessible early-career
46%
carry an explicitly senior title
8
orgs with a machine-readable public feed
How to read this. "Entry-accessible" means the title carries an entry marker (intern, fellow, junior, new-grad…) or the description states a minimum of ≤2 years' experience. It is a measure of advertised, public openings — not the field's true need. Much senior and network hiring never gets posted, so if anything this understates how hard early entry is.

Across 8 mission-driven AI-safety orgs, explicitly senior titles (Chief of Staff, Director, Principal, Head of…) outnumber entry-marked roles many times over. Most remaining titles give no seniority signal at all.

The skew isn't one organization — it holds across the orgs in the sample.

Where a number is named, the floor is high

Most AIS orgs state requirements qualitatively, so only 11 of 69 roles name an explicit minimum — but where they do, the median floor is 4 years, and barely any sit at the ≤2-year level that's reachable early-career. This is a corroborating cross-check on the title signal above, not a stand-alone claim (the sample is small and template-driven).

Frontier labs look the same

Anthropic's public board — kept separate because most of its roles aren't safety work — shows a near-identical early-career share, suggesting the cliff is a field-wide pattern rather than a small-org artifact.

What this is, and isn't

This is a reproducible, source-direct snapshot: every role is pulled from the organization's own hiring system, not a hand-curated job board. That removes the selection bias baked into curated boards — but it covers only orgs that expose a public feed. See Coverage for exactly who is in and out, and Method for how roles are classified and where the limits are.