Safety Research Scholar - Human Frontier Labs

Written by the T&S Network from Scale AI's posting.

Empirical AI safety research on frontier models, run as mentor-guided projects in a small scholar cohort with Scale Labs. Work covers finding model failures, testing whether safety controls hold under pressure, and studying how AI agents misbehave over long tasks. Research areas include AI control, alignment auditing, multi-agent safety, recursive self-improvement safety, evaluation science, and misuse evaluations.

• AI misuse and jailbreaks

What you'd do

• Shape a broad safety concern into testable hypotheses, comparison baselines, and controlled experiments
• Build evaluation harnesses, agent environments, datasets, and monitoring methods
• Weigh confounds and tell apart a model's capability from its typical behavior
• Interpret negative results and choose the next experiment without step-by-step direction
• Meet weekly with a Scale researcher for direction and feedback
• Contribute to papers submitted to top ML conferences such as NeurIPS, ICML, and ICLR

What it asks for

• Experience with technical safety research such as alignment evaluations, AI control, monitoring, or open-weight model safety
• Ability to implement, debug, and interpret experiments with high-level guidance
• Clear research communication and responsible handling of confidential information

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on Scale AInorth_east before you apply. If we got something wrong, report this listing.