Research Program Manager, Safety Systems
About the Team
Safety Systems manages the complete lifecycle of safety efforts for OpenAI’s frontier models. We work across research, engineering, product, and policy to understand risks, develop and validate safeguards, and support responsible deployment.
Our work spans safeguards research and deployment, safety assessments, launch readiness, and trusted-access programs for advanced capabilities. We address risks including biological and cyber misuse and misalignment, with programs that strengthen safety across models and products.
About the Role
We’re hiring Research Program Managers across AI Safeguards and Frontier Risks Safety. You’ll lead programs that develop and improve the safeguards used across our models and products, or assess and mitigate risks associated with frontier capabilities, launches, and trusted-access programs.
You will partner with researchers, engineers, and other stakeholders to turn open safety questions into clear programs of work. You’ll shape technical roadmaps, coordinate execution, and use evidence to guide priorities and inform development and deployment decisions. We’re looking for people who combine strong technical judgment, experience managing complex research programs, and a commitment to AI safety.
These roles are based in San Francisco, CA. We use a hybrid work model of 3 days in the office per week and offer relocation assistance to new employees.
In this role, you will:
- Shape safety programs and technical roadmaps. Partner with research and engineering leads to define goals, milestones, dependencies, and success criteria. Prioritize intake and emerging findings, resolve blockers, and balance immediate safety needs with longer-term research and platform investments.
- Drive safeguards development and deployment. Translate safety research into safeguards across models and products. Maintain an end-to-end view of safeguard coverage, effectiveness, and opportunities for improvement for one or more risk areas, and drive a roadmap that advances across launches.
- Coordinate evaluations and safety assessments. Organize capability evaluations, alignment and control assessments, adversarial testing, and safeguard stress-testing, including risks from autonomous and multi-agent systems. Support embedded assessments throughout model development, internal use, and deployment.
- Drive safety readiness and trusted-access programs. Coordinate safety requirements and execution across frontier model launches and deployments. Work with technical teams to define assessment scope and the evidence needed to inform safety and access decisions.
- Turn findings into improvements. Triage safeguard gaps, jailbreaks, and overflagging with accountable teams. Establish owners and escalation paths, track remediation and regression testing, and turn recurring issues into research and engineering improvements.
- Strengthen the evidence behind safety decisions. Support safety cases that connect claims to evaluations, safeguards, assumptions, and remaining uncertainty. Surface evidence gaps and unresolved risks, and communicate methods, results, limitations, and follow-up actions through assessment reports, system cards, and updates to decision-makers.
- Build programs that scale across products. Coordinate shared tooling and feedback workflows, privacy and security requirements, and safeguard user experience and quality assurance. Help teams apply safeguards consistently across product surfaces and upcoming launches.
You might thrive in this role if you:
- Have 5+ years of experience in technical safety work, owning technical programs, or related roles in fast-moving environments.
- Have led complex research or technical programs involving researchers and engineers, particularly in AI, cybersecurity, biosafety, or another field where failures can have serious consequences.
- Understand frontier AI safety concepts, including capability evaluations, alignment, control, monitoring, and safeguard robustness, and can engage substantively with technical teams.
- Can assess what evaluation results do and do not establish, including methodological limitations, uncertainty, and the difference between model capability and real-world risk.
- Can turn ambiguous questions into clear programs, make practical tradeoffs, and adjust priorities as evidence changes.
- Have worked extensively across research, engineering, and product teams and can coordinate with policy and communications partners.
- Exercise sound judgment when findings are concerning or incomplete, raise unresolved risks constructively, and communicate complex technical work clearly.
- Are motivated to help ensure increasingly capable AI systems remain beneficial and under meaningful human control.