Role overview

The Human Frontier Collective, in partnership with Scale Labs, is seeking Safety Research Scholars to conduct empirical research on artificial intelligence safety. This role involves investigating critical questions about how advanced AI systems fail, whether safety mechanisms withstand real-world pressures, and how AI agents behave in complex, long-duration tasks. Scholars will contribute to foundational research that strengthens the evaluation and safeguarding of frontier AI systems.

Key responsibilities

Own a substantial research project from initial safety question through hypothesis development, baseline design, and controlled experimentation.
Build measurement systems including evaluation harnesses, agent environments, datasets, and monitoring methods tailored to your research focus.
Exercise independent research judgment to identify confounding variables, interpret results, manage negative outcomes, and determine subsequent experiments with high-level guidance.
Implement and debug code, analysis, and experimental infrastructure while maintaining reproducibility standards.
Produce technical writing and research papers suitable for submission to top-tier machine learning conferences such as NeurIPS, ICML, and ICLR.
Meet weekly with a Scale researcher for direction-setting and feedback while driving day-to-day work progress.

Research areas

AI control and chain-of-thought monitoring, including detection of harmful intent or deception.
Alignment auditing and evaluations to identify hidden goals, misaligned behaviors, and unsafe tendencies.
Multi-agent safety, studying risks from agent interactions such as collusion and coordinated failures.
Recursive Self-Improvement (RSI) safety to maintain alignment as AI systems contribute to their own development.
Science of Evaluations, examining model awareness of testing and effects on evaluation reliability.
Misuse risks and evaluations, assessing AI exploitation potential by malicious actors.

Requirements

Pursuing or having completed a PhD in computer science, machine learning, statistics, mathematics, or a related field, or demonstrating an equivalent research track record.
Motivation to reduce catastrophic risks from advanced AI systems and genuine interest in technical safety research.
Demonstrated ability to make progress independently with high-level guidance, including problem-solving, debugging, interpreting results, and proposing experiments.
Skill in implementing and managing research code, experiments, and analysis with attention to reproducibility.
Ability to communicate research clearly and handle confidential information responsibly.

What makes this role worth considering

This opportunity allows you to work on some of the most pressing questions in AI safety within a dedicated research collective. You’ll have direct access to mentorship from Scale researchers while maintaining substantial autonomy over your project direction and execution, creating an ideal environment for scholars seeking to develop independent research capabilities on consequential problems.