Research Engineering Manager, Responsible Frontier AI Research, DeepMind

Overview

As a Research Engineering Manager at DeepMind, you lead a team focused on responsible frontier AI — aligning research, engineering, and policy to build scalable safety and evaluation platforms. You balance research experiments with robust infrastructure, guiding cross‑functional collaboration with researchers and product partners. You shape long‑term technical roadmaps for evaluation pipelines, red‑teaming tools, autoreviewers, and mitigation workflows as frontier models evolve. You will present trade‑offs to senior leadership and drive high‑impact safety initiatives.

Responsibilities

• Set priorities across multiple workstreams to support organizational goals
• Align strategy, processes, and decision‑making across teams
• Move engineering capacity as priorities change across research projects
• Set expectations with tech leads based on level and role
• Conduct regular performance discussions, provide coaching, and lead hiring to build a high‑performing team
• Develop long‑term technical goals and roadmaps for evaluation pipelines, red‑teaming tools, autoraters, and mitigation training pipelines
• Evolve strategy as frontier models become more capable, agentic, and multimodal
• Oversee system designs and code reviews to solve ambiguous problems
• Set best practices for code quality, scalability, reliability, and secure data handling
• Partner with Research Scientists, Subject‑Matter Experts, Product, Program, Policy, and Trust & Safety to translate research into evaluations and mitigations for model launches
• Present trade‑offs and results to senior leadership

Key requirements

• Experience leading teams working on AI safety, security, or responsible AI research (e.g., adversarial robustness, red teaming, safety evaluations, or mitigations for frontier models)
• Experience with automated red teaming or jailbreak discovery, LLM‑based autoraters and evaluation design, safety classifiers or probes, prompt injection and agent security, or evaluation of severe‑harm risks such as CBRN or Harmful Manipulation

• leadership and people management
• communication with senior leadership
• cross‑functional collaboration
• automated red teaming or jailbreak discovery
• LLM‑based autoraters and evaluation design
• safety classifiers or probes

…