Researcher, Agent Safety, Oversight and System Mitigations

Highlights

• Build oversight systems like agent-based review for real-world Codex deployments.
• Red-team agentic systems to test controls against exfiltration and unsafe tool use.
• Improve safety-productivity tradeoff by cutting missed harms and approval burden.

What they offer

• Equity
• Relocation assistance

Ideal candidate

Safety-minded researcher or engineer with strong systems instincts who enjoys turning ambiguous safety questions into tested mitigations and cares deeply about frontier AI alignment.

Requirements

• Strong systems or security instincts for reasoning about isolation, permissions and failure modes
• Turn ambiguous safety questions into threat models, experiments and practical mitigations
• Build robust experimental infrastructure and evaluations for safety mitigations
• Deep interest in frontier AI alignment, safety and control