JOB DESCRIPTION

The AI Security Institute (AISI) is seeking researchers to join its newly formed Agentic AI Risk Modelling and Mitigations team in London. This team will develop rigorous threat models and practical mitigation strategies to address the growing risks posed by advanced, autonomous AI systems.

As AI systems become more capable and agentic, understanding how loss-of-control scenarios could emerge is critical. This role focuses on building structured, defensible risk models that identify plausible harm pathways and translate them into actionable government interventions. The work bridges AI safety research, cybersecurity, national security, and public policy.

You will collaborate with AISI red teams, evaluation teams, alignment researchers, and UK government stakeholders. The role involves analyzing empirical evaluation data, drawing from cybersecurity and machine learning literature, and developing precise models of agentic AI risks such as deceptive alignment, long-horizon cyber operations, or misuse scenarios.

Responsibilities include developing detailed threat models, translating risk pathways into evaluation specifications, analyzing mitigation effectiveness (including monitoring infrastructure and governance mechanisms), and communicating findings to both technical and policy audiences. Some projects may include hands-on ML or cybersecurity collaboration to prototype mitigations.

Candidates may be hired at junior through principal research levels, with leadership opportunities available for highly experienced applicants.

Required qualifications include experience producing structured analytical work such as threat models or safety cases, strong written communication skills, familiarity with cybersecurity or AI safety literature, and the ability to reason rigorously about uncertain, high-impact scenarios.

Preferred qualifications include hands-on LLM experience, familiarity with AI capability evaluations, mentoring experience, and eligibility for UK government security clearance.

This position is UK-based, requires working at least 9 days per fortnight, and involves at least 3 days per week onsite in London.