Research Scientist, Agentic Safety

As a Research Scientist in Strategic Initiatives, you will use your machine learning expertise to collaborate with other machine learning scientists and engineers within our strategic initiatives programs. Your primary focus will be on building technologies to make AI agents safer. AI agents are increasingly used in sensitive contexts with powerful capabilities, having abilities to access personal data, confidential enterprise data and code, interact with third party applications or websites, or write and execute code in order to fulfil user tasks. Ensuring that such agents are reliable, secure and trustworthy is a  large scientific and engineering challenge, with huge potential impact. In this role, you will serve this mission by proposing and evaluating novel approaches to agentic safety, building prototype implementations and production grade systems to validate and ship your ideas, in collaboration with a team of researchers and engineers from SSI, and the rest of Google and GDM.

Key responsibilities:

• Invent and implement novel recipes for making agents safer, involving both improving models that power the agents, as well as systems that are built around the agents
• Develop strategies to hill-climb leaderboards and debug possible performance and safety issues in frontier agents
• Integrate novel agentic technologies into research & production grade prototypes 
• Work with product teams to gather research requirements and consult on the deployment of research-based solutions to help deliver value incrementally
• Amplify the impact by generalizing solutions into reusable libraries and frameworks for safer AI agents across Google, and by sharing knowledge through design docs, open source, or external blog posts