Anthropic Curated from employer
Cyber Evaluations Engineer
United States · Remote-friendly · San Francisco / Washington, DC
USD $300,000–$405,000 / yearView role Source checked 2026-10-30
SPECIALIST CAREERS · CANADA & THE US
AI safety engineer jobs
Build and evaluate systems that help teams understand and reduce AI risks. Compare safety tooling, safeguards and adversarial evaluation roles.
Safety toolingAdversarial testingPython & software systemsRisk evaluation
APPLY DIRECTLY WITH EMPLOYERS
6 current roles
Anthropic Curated from employer
Research Engineer, Post-Training Model Evaluations
San Francisco / Seattle · Hybrid
USD $500,000–$850,000 / yearView role Source checked 2026-10-30
USD $207,000–$385,000 / yearView role Source checked 2026-10-30
Decagon Curated from employer
Research Engineer, Safety
San Francisco / New York City · In-office · Full-time
USD $200,000–$400,000 / yearView role Source checked 2026-10-30
fal Curated from employer
Machine Learning Engineer, Safety
San Francisco · On-site five days a week · Full-time
USD $180,000–$250,000 / yearView role Source checked 2026-10-30
Apollo Research Curated from employer
Research Scientist/Engineer (Evaluations)
San Francisco · In-person · London option also listed · Full-time
Pay: see employerView role Source checked 2026-10-30
No matching roles right now.
Try another skill or remove a filter. We’re starting with a small, focused selection.
Each role shows its source-check date and known location restrictions. A listing can close before its review window ends; confirm current availability with the employer. How we review listings.
UNDERSTAND THE WORK
AI safety
career guide.
By TaskBull AIJobs · Updated October 30, 2026
AI safety engineering includes several kinds of work
Safety engineering can mean building internal tools, assessing model behavior or implementing safeguards around a product. Some roles focus on software used by safety teams; others investigate misuse or test security boundaries. These are related responsibilities, but they require different evidence. Read the problem the team is trying to solve and the kind of system you would build or evaluate.
Distinguish safeguards, evaluation and research positions
An AI safety software engineer can need backend, frontend and database experience. An adversarial evaluation role may emphasize Python, security knowledge and careful experiments. A research vacancy can require a different background again. Do not assume that every safety job requires a doctorate or that all safety work is red teaming. Compare the employer’s required skills, outputs and seniority.
Prepare evidence of careful testing and clear reporting
A useful portfolio can document a small evaluation suite, its threat assumptions and the limits of its results. Explain how you reproduce a failure, avoid misleading metrics and communicate uncertainty. For tooling work, show a maintainable system that helps someone investigate or review cases. Use public or synthetic data and explain what a test demonstrates without claiming it proves a model is safe in every setting.
Check location, access and working expectations
Safety and security teams can have office attendance, sensitive-system access or travel requirements. A source that says remote-friendly does not necessarily allow fully remote work from any country. Confirm the listed location and attendance expectations. Published compensation is specific to the vacancy and location, and TaskBull’s review-window date is not the employer’s application deadline.
Sources and reading
Career guidance draws on the examples below and TaskBull’s editorial analysis. Employer sources describe their own roles; they do not endorse this board.
BEFORE YOU APPLY
Your career
questions.
Are AI safety jobs and AI security jobs interchangeable?
They can overlap in misuse testing and safeguards, but infrastructure security, product security and model safety are different scopes. Use the duties to decide whether a vacancy matches your background.
Do all AI safety engineer jobs involve red teaming?
No. Safety roles can build investigation tools, monitoring or safeguards as well as adversarial tests. We include related evaluation roles only when their responsibilities support the category.