Frontier AI Safety Researcher

Rozgoo; San Francisco (CA); Hybrid; USD 140,000 - 190,000; Full time

Job summary

OpenAI is seeking a Researcher, Safety to advance Frontier Assurance. You will design rigorous assessments of frontier AI risks, collaborate with external evaluators and internal teams, and help translate research questions into experiments, analyses, and evidence requirements.

You will build tools to automate auditing and ensure reproducible results. You will work on designing secure assessment environments and interpreting model behaviors, with a focus on transparency, safety, and rigorous

Qualifications

• Experience designing evaluations for frontier AI risks and safety claims.
• Ability to work with external evaluators and internal teams on rigorous assessments.
• Familiarity with securing assessment environments and reproducible results.

Responsibilities

• Design and implement assessments with external evaluators and internal teams.
• Support third-party evaluations and embedded audits focused on frontier risks.
• Collaborate with external researchers to provide data, documentation, and context for assessments.
• Develop tooling to automate auditing, evidence collection, and evaluation workflows.
• Document methods and results so evaluators can scrutinize evidence and limitations.

Skills

experimental design

machine learning evaluation

assessment automation

debugging complex systems

evaluation infrastructure

secure access controls

reproducible research

safety evaluations

ai safety evaluations

adversarial testing

security research

technical auditing

frontier ai risks

alignment evaluation

Job description

OpenAI is seeking a Researcher, Safety to advance Frontier Assurance. You will design rigorous assessments of frontier AI risks, collaborate with external evaluators and internal teams, and help translate research questions into experiments, analyses, and evidence requirements.

You will build tools to automate auditing and ensure reproducible results. You will work on designing secure assessment environments and interpreting model behaviors, with a focus on transparency, safety, and rigorous