Mercor AI Safety Fund Grants
San Francisco • Temporary
Posted 3h ago
Job Location
San Francisco
Tech Stack
Remote Work Policy
On-site
Employment Type
Temporary
Categories
AI Research Engineer
About the job
Mercor is offering $5 million in grants to fund safety research for frontier AI models. The goal is to address critical challenges in AI safety, ensuring models are safe for deployment beyond controlled environments. This initiative supports research into areas such as misalignment, sandbox escape, evaluation awareness, interpretability, oversight and control, and red-teaming methodologies. Mercor aims to foster advancements in AI safety by providing funding for researcher hours, API credits, and event attendance, and by offering access to their expert network and internal evaluation infrastructure.
Responsibilities
- Conduct research on AI safety, focusing on areas like misalignment, sandbox escape, evaluation awareness, interpretability, oversight and control, and red-teaming.
- Develop robust systems for uncovering novel AI failures.
- Publish research findings in the form of papers, open datasets, public methodologies, or tools.
- Collaborate with Mercor's expert network for human grading, red-teaming, and annotation, if applicable to the research.
- Utilize Mercor's internal evaluation infrastructure, subject to review.
Requirements
- Independent researchers, academics, PhD students, and small teams are encouraged to apply.
- Applicants should have a specific, well-scoped research question.
- A background in ML, CS, statistics, or an adjacent field (measurement, psychometrics, HCI, security, social science) is required.
- Experience with agentic evaluation, RL environments, adversarial ML, or systems security is a bonus.
- Ability to publish research outcomes (paper, dataset, methodology, or tool).
Benefits
- Funding for researcher time
- API credits
- Stipends for event and conference attendance
- Access to Mercor's expert network for human grading, red-teaming, and annotation
- Access to Mercor's internal evaluation infrastructure