Research Engineer, Frontier Safety Mitigations

About the job

The goal of our frontier safety mitigations work is to de-risk model launches by researching and implementing defenses against high-stakes frontier safety risks, particularly those coming from misuse that could make a model tangibly dangerous as model capabilities increase. In this role, you will be accountable for the safety and behavior of Google DeepMind’s (GDM) latest Gemini models. You will focus on critical domains such as CBRN (Chemical, Biological, Radiological, Nuclear), Cybersecurity, and Harmful Manipulation and will make sure that our mitigations are still enabling the beneficial use of our technology. You will employ a wide range of methods, from building novel evaluations to red-teaming, researching and deploying advanced mitigations, monitoring emerging risks, and contributing to model development.

Responsibilities

• Develop high quality evaluations that capture risks arising in frontier safety domains, such as cybersecurity and CBRN.
• Design, implement, and productionalize a range of safety mitigations, including in-model approaches (e.g., Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) training recipes) and out-of-model solutions (e.g., logging, monitoring).
• Own the data and evaluation pipeline to measure the effectiveness of our mitigations.

Minimum qualifications:

• Bachelor’s degree or equivalent practical experience.
• 5 years of experience with software development in one or more programming languages.
• 3 years of experience testing, maintaining, or launching software products, and 1 year of experience with software design and architecture.

Preferred qualifications:

• Master's degree or PhD in Computer Science or a related technical field.
• Experience in Large Language Model (LLM) development, fine-tuning, or safety evaluation methodologies.
• Experience taking research from concept to product.
• Experience in areas such as Safety and Alignment.
• Familiarity with concepts like the Frontier Safety Framework, adversarial attacks, and in-model/out-of-model mitigation strategies.
• Ability to build large-scale research or engineering systems.