AI Safety Career Intelligence 2025-2026

AI Safety Career Paths and Resources

Explore high-impact paths in technical safety, governance, and strategy.

AI safety workforce (2025)

1,100+ FTEs

Growth from ~400 FTEs in 2022 to >1,100 in 2025.

Technical vs. non-technical split

600 / 500 FTEs

Technical safety remains under-scaled relative to capability teams.

Senior private-lab compensation

$500k-$1M+

Top roles can exceed this range when equity is included.

US government technical ceiling

$197,200

Public-sector compensation has adjusted to compete for expert talent.

Private Labs

Anthropic, Google DeepMind, and OpenAI are scaling empirical safety, evals, and alignment engineering.

Public Sector

US and UK AI Safety Institutes are formalizing testing standards, model audits, and regulatory capacity.

Civil Society

Think tanks and non-profits shape strategy, policy analysis, field-building, and talent pipelines.

Four Career Domains

Interactive view

Why this domain matters

The field has shifted from philosophical speculation to empirical engineering: interpretability, robustness, and evals now govern deployment decisions.

Role landscape

Mechanistic Interpretability Researcher (circuit discovery, feature mapping, universality testing).
Alignment and Robustness Researcher (RLHF, Constitutional AI, scalable oversight, adversarial training).
Model Evaluation Engineer (red-teaming, dangerous capability benchmarks, threat modeling).

Core competencies

Python + PyTorch/JAX fluency.
Strong understanding of Transformer internals (attention, residual streams, layer norms).
Math foundation: linear algebra, calculus, probability.
Portfolio proof via paper replications, eval tooling, or published safety experiments.

Entry pathways

MATS (12-week mentorship; highly selective).
ARENA (4-5 week intensive, with open curriculum).
ML4Good and PIBBSS for foundational and interdisciplinary entry.
Independent track: Zero-to-Hero -> ARENA modules -> capstone on Alignment Forum/LessWrong.

Domain comparison matrix

Domain	Impact Estimated societal risk-reduction leverage of each domain. Higher means stronger direct influence on catastrophic-risk mitigation outcomes. Evidence:	Barrier Entry difficulty based on typical credential depth, technical specialization, and hiring selectivity. Higher means harder entry. Evidence:	Time to readiness Approximate months/years needed to become competitive in that domain, informed by fellowship lengths, curriculum depth, and expected portfolio evidence. Evidence:	Compensation Relative earnings potential normalized from sector ranges (industry, public sector, nonprofit, specialist security roles). Evidence:	Policy leverage Ability for work in the domain to shape standards, regulation, and governance enforcement at organizational or national scale. Evidence:
Technical AI Safety Technical AI Safety High impact and compensation reflect frontier-lab demand and tight specialist supply; policy leverage is lower because influence is mostly technical, not regulatory. Evidence:	95	88	70	94	60
Specialized Intersections (Bio, Cyber, Law) Specialized Intersections (Bio, Cyber, Law) Scores are high because dual-domain talent (bio/cyber/law + AI) is scarce and high leverage, but readiness and barriers increase due to required multidisciplinary depth. Evidence:	90	82	76	84	86
AI Governance & Policy AI Governance & Policy Policy leverage is highest due to direct influence on legislation, standards, and institutional deployment controls; entry barrier is moderate vs deep technical tracks. Evidence:	88	68	62	72	96
Strategy, Operations & Field-Building Strategy, Operations & Field-Building Lower barrier and faster readiness come from transferability of operations/program skills; impact is amplified through execution and coordination rather than direct model control. Evidence:	78	52	55	66	74

Relative scores (0-100) built from evidence-weighted rubric scoring. Use tooltips on headers/rows to inspect assumptions and source basis.

Compensation landscape

Filter by sector and geography to compare min, median, and max compensation ranges.

MedianRange min-max

Industry labs - Entry Research Engineer

$200,000 - $300,000

Median: $250,000

Industry labs - Senior Research Scientist

$500,000 - $1,000,000

Median: $750,000

Think Tank - Junior Researcher

$70,000 - $100,000

Median: $85,000

Think Tank - Senior Fellow / Manager

$120,000 - $180,000

Median: $150,000

AI Security - Specialist Roles

$125,000 - $320,000

Median: $180,000

Geographic hubs

San Francisco Bay Area
Highest density of frontier labs and technical safety talent.
London
Strong technical + governance concentration (DeepMind, GovAI, UK AISI).
Washington, D.C.
US policy and regulatory center for federal AI governance careers.
Beijing
Growing safety/governance ecosystem around BAAI and related institutions.

Programs and fellowships

MATS

Mentorship-based technical research

12-week program; competitive admission (~4-7%) and close mentor matching.

ARENA

Technical upskilling

4-5 week intensive with practical RL/Transformer/evals track and open curriculum.

Horizon Fellowship

US federal policy placement

AI and biosecurity pathways into executive branch, Congress, and policy institutions.

TechCongress

Legislative advising

Places technologists directly in US Congressional offices.

GovAI Fellowships

Research / policy / operations tracks

Oxford/London ecosystem with structured fellow pathways.

AAAS Fellowships

Science policy entry

Large placement engine for PhD-level talent in US government.

Immediate action plan

1Read foundation guides
Start with 80,000 Hours career materials and CAIS-style intro pathways.
2Build network surface area
Join AI Alignment / EA communities and engage in public technical discussion.
3Pick one high-signal upskilling path
BlueDot for governance or ARENA/MATS-style tracks for technical depth.
4Ship a proof-of-work artifact
Publish evals, replications, or applied governance memos tied to real model risk.

Shift from theory to empiricism

Alignment work now depends on measurable experiments and eval infrastructure.

Intersections are expanding

Biosecurity, cybersecurity, and legal engineering now require hybrid career profiles.

Operations is leverage

Research outcomes depend on managerial throughput, funding operations, and community infrastructure.

Source explorer

Click any citation token in the page or use the source index below to open details.