Frontier AI Safety Research Engineer

Google LLC; Greater London; Hybrid; GBP 110,000 - 170,000; Full time

Job summary

Google DeepMind in London seeks a highly adaptable Research Engineer to join the Responsibility team. You will design and scale evaluation frameworks that assess frontier language models for harmful manipulation, develop mitigation techniques, and build infrastructure to track model performance across releases.

You will be joining a specialized Applied Research and Responsibility team within Google while closely partnering with DeepMind, Research, Product, and Policy to address AI safety

Qualifications

• Bachelor's degree or equivalent practical experience in CS/ML/math.
• 3 years of experience in Python programming.
• 3 years of experience with ML frameworks (JAX, PyTorch, TensorFlow).

Responsibilities

• Prototype and deliver scalable engineering solutions across the Responsibility portfolio.
• Architect and optimize training and inference pipelines to detect manipulation in frontier models.
• Develop post-training strategies to mitigate manipulation, including deceptive persuasion and covert influence.
• Collaborate with researchers to translate safety research into robust implementations and present results to stakeholders.
• Build and maintain evaluation infrastructure to track model safety performance across releases.

Skills

Python programming

ML frameworks (JAX, PyTorch, TensorRT)

Education

Bachelor's degree in CS/ML/Math or related field

Master's or PhD in CS/Engineering related to ML

Tools

Python

C++

JAX

PyTorch

TensorFlow

Job description

Google DeepMind in London seeks a highly adaptable Research Engineer to join the Responsibility team. You will design and scale evaluation frameworks that assess frontier language models for harmful manipulation, develop mitigation techniques, and build infrastructure to track model performance across releases.

You will be joining a specialized Applied Research and Responsibility team within Google while closely partnering with DeepMind, Research, Product, and Policy to address AI safety