Research Scientist – AI Safety

Overview

In this Research Scientist role at Faculty, you conduct original AI safety research to enable safe, trustworthy deployment. You design and execute impactful experiments, study model behavior, and build robust tooling that accelerates discovery. You collaborate with senior scientists to translate findings for external partners and national security institutes, helping shape safety standards. You will work at the intersection of theory and practice to address critical challenges in modern AI. This is an opportunity to contribute to responsible AI at scale within a fast-growing, research-driven environment.

Pay / Benefits

• Unlimited Annual Leave Policy
• Private healthcare and dental
• Enhanced parental leave
• Family-Friendly Flexibility & Flexible working
• Sanctus Coaching
• Hybrid Working

Responsibilities

• Design and run experiments to study and steer transformer-based LLM behavior
• Apply white-box and black-box interpretability methods, focusing on steering and mechanistic interpretability
• Develop modular, reproducible research code in Python using PyTorch
• Build and maintain research infrastructure and tooling to accelerate discovery
• Document findings through academic papers, reports, and presentations
• Collaborate with senior scientists to refine hypotheses and research scope
• Translate complex insights for external partners, frontier labs, and safety institutes

Key requirements

• Solid background in machine learning fundamentals and transformer architectures (LLMs)
• Practical understanding of interpretability techniques (white-box/black-box) such as steering vectors
• Ability to formulate hypotheses from structured technical problems
• Strong Python programming skills and PyTorch experience; clean, modular code; Git version control
• Excellent communication to translate concepts and results into reports and presentations
• Curiosity and analytical rigor with exposure to safety areas like robustness, evaluation, or uncertainty calibration

• Curiosity
• Analytical rigor
• Strong written and verbal communication
• Python
• PyTorch
• Transformer architectures (LLMs)

…