Security Engineer (AI Red Team) at Refractal
United Kingdom
Posted
on-site · full_time · Visa sponsorship
Apply for this role at Refractal
Responsibilities:
- Design and execute authorised adversarial assessments against deployed AI systems and agent workflows.
- Simulate realistic attacker goals (prompt injection, tool misuse, sensitive-data exfiltration, privilege escalation) within defined test scopes.
- Build reproducible test cases and lightweight tools that demonstrate security impact and separate reliable failures from flukes.
- Analyse how model behaviour, application design, identity and permissions combine to produce vulnerabilities.
- Document findings clearly, recommend practical mitigations, retest fixes with engineering teams, and contribute regression tests.
Requirements:
- Hands-on experience in offensive security, application security testing or adversarial testing of AI/LLM systems.
- Strong understanding of web apps, APIs, authentication and authorization failure modes.
- Knowledge of prompt-based systems: prompts, retrieval/memory, connected tools and trust boundaries.
- Ability to write automation code to reproduce findings, run tests and inspect application behaviour.
- Good judgment around test scope and sensitive data, and strong written/verbal skills to explain practical impact.