Frontier AI Safety Research Scientist

AI Security Institute; Greater London; On-site; GBP 65,000 - 145,000; Full time

Benefits offered by this job

Hybrid working

Pension contribution

Conference funding

Learning stipend

Job summary

AI Security Institute in London is seeking research scientists and engineers to advance misuse research and defenses for frontier AI models. You will lead technical direction or execution, contribute to publications, and work across teams with collaborators from industry and government.

The role offers hybrid work near London with opportunities to shape AI security policy and practice, including publishing and collaboration with frontier AI labs.

Qualifications

• Hands-on research with large language models (LLMs) including training, evaluation, or safety research.
• Demonstrated track record of peer‑reviewed publications in top ML conferences or journals.
• Ability to write clean, documented research code for ML experiments (PyTorch or similar).
• Strong sense of mission and urgency, with ability to work self‑directed while collaborating with others.

Responsibilities

• Design, build, and evaluate methods to attack and evaluate safeguards for frontier AI systems.
• Build benchmarks for monitoring misuse across multiple model interactions.
• Investigate novel attacks and defenses for LLMs and other frontier AI technologies.
• Perform adversarial testing of frontier AI system safeguards and report actionable findings.

Skills

LLMs

Publications

Research code

Mission-driven

Self-directed

Teamwork

Tools

PyTorch

Inspect

Job description

AI Security Institute in London is seeking research scientists and engineers to advance misuse research and defenses for frontier AI models. You will lead technical direction or execution, contribute to publications, and work across teams with collaborators from industry and government.

The role offers hybrid work near London with opportunities to shape AI security policy and practice, including publishing and collaboration with frontier AI labs.