Research Scientist, Gemini Safety & Behavior — GenAI

Google LLC; New York, Mountain View (NY, CA); On-site; USD 207,000 - 300,000; Full time

Benefits offered by this job

Equity

Bonus target

Benefits

Job summary

DeepMind is seeking a Research Scientist for Gemini Safety and Behavior to advance safety-oriented research from concept to live release. You will work within a fast-paced, highly collaborative team focused on auditing frameworks, defensive safeguards, and autonomous agents for flagship systems used by billions of users.

The role emphasizes engineering novel heuristic and statistical mechanisms, rapid incident mitigation, and close collaboration with product areas to scale safety solutions

Qualifications

• PhD degree in Computer Science, a related field, or equivalent practical experience.
• 2 years of experience in Large Language Model safety or security.

Responsibilities

• Drive innovation and understanding of safety and security among Gemini models, delivering mitigations at scale.
• Develop red and blue teaming methods for frontier GenAI models, including text, multimodal, and agentic capabilities.
• Explore data, reasoning and algorithmic solutions to ensure Gemini models are safe and maximally helpful.
• Improve Gemini’s adversarial robustness focusing on high-stakes abuse risks in agentic settings.
• Develop and execute experimental plans to address gaps or create new capabilities.

Skills

Agentic workflows

Safety and alignment

Synthetic data pipelines

Concept-to-product research

Applied research leadership

Experimentation & baselines

Publications (NeurIPS/ICML/ICLR)

Education

PhD in CS or related field or equivalent

Job description

DeepMind is seeking a Research Scientist for Gemini Safety and Behavior to advance safety-oriented research from concept to live release. You will work within a fast-paced, highly collaborative team focused on auditing frameworks, defensive safeguards, and autonomous agents for flagship systems used by billions of users.

The role emphasizes engineering novel heuristic and statistical mechanisms, rapid incident mitigation, and close collaboration with product areas to scale safety solutions