Requirements

• Bachelor degree or equivalent practical experience in a management, scientific, technical or engineering field.
• 8 years of experience writing code in a shared, version-controlled codebase with other contributors, which may include research software.
• 3 years of experience training, fine-tuning or evaluating machine learning models.

Nice to haves

• Master's degree or PhD in Chemistry, Physics, Chemical or Nuclear Engineering, Biology, or Computational Science, or a related field.
• Experience building classifiers or evaluation infrastructure for large language models.
• Experience applying statistics to machine learning evaluation.
• Experience running software in production, including monitoring, on-call or incident response.
• Experience in chemical, biological, radiological or nuclear security, export control, or adversarial work such as red teaming, security research or abuse prevention.

What you'll be doing

• Build novel safeguards and attack them with automated and manual red teaming, while measuring their cost to legitimate scientific research.
• Design, build and run evaluations for chemical, biological, radiological, nuclear and explosives risk in frontier models, including agentic systems risk.
• Architect the reliability of key systems, including safeguards and incident response.
• Partner with leading scientific experts to turn their judgment into datasets and automated graders, managing sensitive data under strict security controls.
• Communicate results, including uncertainty and negative findings, to technical staff and leadership.

Perks and benefits

• Opportunity to work in Google DeepMind, a pioneering AI lab focused on advancing AI development for global challenges.
• Collaborative interdisciplinary teams including biologists, chemists, and physicists.
• Impactful role contributing to safety and ethics in frontier AI research.