AI Safety Jobs

A role focused on ensuring advanced AI systems behave as intended and avoid harm, spanning alignment, red-teaming and evaluations, safety policy, and pre-deployment governance.

Primarily at frontier AI labs and organizations building or deploying the most capable models, with a mix of technical research and governance work.

Less GRC tooling and more evaluation infrastructure than the compliance side. Model evaluation and red-teaming harnesses, incident and near-miss logging, capability and dangerous-capability test suites, experiment tracking, and the internal documentation systems that record what was tested and what the results were. Governance tooling still appears where a safety case has to be written down and defended, particularly at labs operating under a responsible scaling policy.

It splits into two hiring tracks. The technical track wants machine learning depth, evaluation design, red-teaming and the engineering to build test harnesses. The governance and policy track wants risk analysis, framework fluency, technical writing and the ability to translate an evaluation result into a decision a leadership team can act on. Both need unusually careful reasoning about failure modes that have not happened yet.

Design an evaluation, run it against a model, interpret ambiguous results, and write up what it does and does not tell you. Feed that into a decision about whether a capability ships, ships with mitigations, or does not ship. Log incidents and near misses, revise the test suite as capabilities change, and defend the safety case to people who want a cleaner answer than the evidence supports.

For the technical research roles, usually yes. For the growing governance, policy, evaluation-operations and program roles around them, no. Those hire from risk, compliance, audit and policy backgrounds, and they are where most of the recent job growth has been. The technical bar is real but it is literacy, not a doctorate.