Safety Research Scholar - Human Frontier Labs

Scale AI

4h

About The Position

The Safety Research Scholar program brings researchers together to investigate the empirical foundations of AI safety. Scholars own a substantial portion of a safety research project, working with a mentor from formulating a research question through experiments and deciding what to test next. Projects are matched to demonstrated expertise, research interests, and background; scholars are not expected to cover every research area. By default, scholars receive a list of proposals and are matched with one. They meet weekly with a Scale researcher who sets direction and provides feedback, while the scholar drives day-to-day work.

Research areas include AI control and chain-of-thought monitoring; alignment auditing and evaluations; multi-agent safety; recursive self-improvement safety; the science of evaluations; and misuse risks and evaluations. Work may involve turning high-level safety questions into hypotheses, baselines, and controlled experiments; implementing evaluation harnesses, agent environments, datasets, monitoring methods, or post-training experiments; exercising independent research judgment; and producing reproducible code, experiment logs, analysis, and precise technical writing. Outputs may include research papers submitted to top ML conferences such as NeurIPS, ICML, and ICLR. External outputs are subject to research, security, and communications review where appropriate.

Requirements

• Motivation to reduce catastrophic risks from advanced AI systems.
• Interest in technical AI safety research, such as alignment evaluations, AI control, monitoring, multi-agent safety, or open-weight model safety.
• Pursuing or having completed a PhD in computer science, machine learning, statistics, mathematics, or a related field, or having an equivalent track record.
• Ability to make progress with high-level guidance, including implementing and debugging work, getting unstuck, interpreting results, and proposing next experiments.
• Ability to communicate research clearly.
• Ability to handle confidential information and internal systems responsibly.

Responsibilities

• Own a safety research project from a high-level question through hypotheses, baselines, controlled experiments, and research outputs.
• Implement and debug evaluation harnesses, agent environments, datasets, monitoring methods, or post-training experiments.
• Identify confounds, distinguish model capabilities from typical behavior, interpret negative results, and propose subsequent experiments with high-level guidance.
• Produce reproducible code, experiment logs, analysis, and precise technical writing.
• Meet weekly with a Scale researcher for research direction and feedback, while driving day-to-day project work.

What This Job Offers

• Job Type: Full-time
• Career Level: Mid Level
• Education Level: Ph.D. or professional degree