Red Team Specialist - Cyber
OpenAI · San Francisco
About this role
**Red Team Specialist - Cyber** **About the Team** The Intelligence and Investigations team ensures safe, responsible AI deployment by rapidly detecting and mitigating abuse. We leverage advanced testing methodologies to uncover vulnerabilities and emerging threats, safeguarding OpenAI's products and users. **About the Role** Help answer critical questions: What cyber capabilities can our models provide to real-world attackers, and do our safeguards remain effective against increasingly sophisticated techniques? This role combines scaled evaluation with expert-driven testing, requiring strength in either cybersecurity or model evaluation with foundational fluency in both areas. **Key Responsibilities** • Design and run rigorous evaluations of model cyber capabilities and safeguards • Conduct hands-on testing to understand model capabilities when used by security practitioners • Distinguish benchmark failures from meaningful real-world risks • Build and improve automated testing infrastructure for repeatable measurement • Test novel abuse risks in agentic systems (prompt injection, agent hijacking, etc.) • Translate findings into clear risk assessments and actionable recommendations • Contribute to Safety Bug Bounty work requiring cyber expertise **Ideal Candidate** • Substantial depth in cybersecurity (penetration testing, vulnerability research, red-team ops) OR AI model evaluation (evals design, agentic harnesses, adversarial testing) • Working literacy across both cybersecurity and model evaluation • Strong coding ability for building testing tools and automating experiments • Attacker mindset with interest in discovering failure modes • Clear communication across technical and non-technical audiences • Experience translating findings into decisions and mitigations **Location & Work Model** San Francisco, CA or Seattle, WA | Hybrid (3 days/week in office) | Relocation assistance available
Listing freshness
CronJobs last confirmed this listing 15h ago. If its source stops confirming the opening for seven days, this page is removed from active inventory.