Red-Team Engineer, Adversarial Robustness

Role overview Test AI agents and related systems by finding ways their instructions, tools, or permissions can be abused. You’ll build repeatable attack suites and regression checks so teams can understand real risks, apply fixes, and verify that later releases have not reopened the same vulnerabilities.

Responsibilities - Design and run attacks involving prompt injection, privilege escalation, goal hijacking, and misuse of tools. - Track vulnerabilities across releases and check whether fixes introduce related weaknesses. - Write reproducible findings with impact-based severity assessments and practical mitigations. - Adapt emerging attack techniques to the specific tools, permissions, and trust boundaries of the systems under review. - Test systems developed by the engineering team before deployment.

Requirements - Experience red-teaming LLM or agentic systems, or application security experience with the ability to learn agent-specific attack surfaces. - Ability to create working exploits and document clear reproduction and remediation steps. - Familiarity with agent frameworks, tool-calling patterns, and the difference between documented and actual trust boundaries. - Sound judgment in assessing severity and distinguishing meaningful risk from low-impact findings.