Research Engineer, Responsible Frontier AI Research, DeepMind

Minimum qualifications:

• Bachelor's degree in Computer Science, Machine Learning, Mathematics, or a related technical field, or equivalent practical experience.
• 3 years of experience in Python programming.
• 3 years of experience with ML frameworks such as JAX, PyTorch, or TensorFlow.

Preferred qualifications:

• Master's degree or PhD in Computer Science, Engineering, or a related field with a focus on machine learning.
• Experience in Python and C++ for high-performance ML library development.
• Experience with harmful manipulation detection, persuasion modeling, deceptive behavior analysis, or AI safety evaluation and mitigation.
• Experience working directly on AI safety, or responsible AI research.
• Experience building evaluation frameworks, benchmarks, or automated testing pipelines for ML models.

About the job

We are looking for a highly adaptable Research Engineer to join our Responsibility team. In this role, you will design and scale evaluation frameworks that assess the propensity of frontier language models to engage in harmful manipulation, develop evaluation and mitigation techniques for manipulative model behaviors, and build infrastructure to systematically track model performance across releases. You will work closely with Research Scientists, safety policy teams, and product stakeholders to ensure that evaluation results translate into concrete safety improvements.

You will be joining a specialized Applied Research and Responsibility team within Google while closely partnering with teams across Deepmind, Research, Product, and Policy to identify and address key challenges in responsible AI. Our work remains deeply integrated with DeepMind.

Responsibilities

• Be able to rapidly prototype and deliver scalable engineering solutions across the Responsibility research portfolio.
• Architect and optimize training and inference pipelines to detect and evaluate harmful manipulation behaviors in frontier language models.
• Develop post-training strategies to mitigate manipulation risks including deceptive persuasion, sycophancy, and covert influence tactics.
• Collaborate with research scientists to translate safety research into robust implementations and present results to cross-functional stakeholders.
• Build and maintain evaluation infrastructure to systematically track model safety performance across releases.