Research Scientist, Safety Oversight, DeepMind

Overview

In this role you will build classifiers and data pipelines to detect model misbehavior and misuse end-to-end, and develop cross-context monitoring to identify large-scale attack vectors. You will explore novel monitoring methods using model activations, actions, and final outputs, collaborating with infrastructure and data science teams to scale results. You’ll contribute to reporting and evaluating safety signals across production AI systems, supporting rapid mitigation of safety risks. This position sits in the GenAI Safety organization with a focus on proactive production-time oversight and safety validation.

Responsibilities

• Develop classifiers and data pipelines to detect misbehavior and misuse of deployed models end-to-end
• Research cross-context monitoring and novel signal aggregation for large-scale attack detection
• Explore monitoring methods leveraging model activations, actions, chains-of-thought, and final answers
• Collaborate with infrastructure teams and data scientists to scale monitoring solutions
• Regularly share results with the wider safety team and contribute to safety validation in production

Key requirements

• PhD in Computer Science, a related field, or equivalent practical experience
• Experience in generative AI and Large Language Models (LLM)
• Experience building and shipping technical products
• Ability to work in fast-paced, collaborative environments and communicate results clearly
• Experience developing data pipelines and automated evaluation designs

• strong collaboration
• analytical mindset
• problem-solving initiative
• building highly parallelised data pipelines
• data quality and automated evaluation design
• statistical modeling

…