Overview In this role you will help build classifiers and end-to-end data pipelines to detect model misbehavior and misuse. You will research cross-context monitoring systems to identify coordinated harms and develop novel signal aggregation methods across user sessions. You will think critically about monitoring using model activations, actions, chains-of-thought, and final outputs. You will collaborate with infrastructure and data science teams to scale your work and share results with the safety team to improve deployed model safety and real-world effectiveness.
Compensation / Benefits- bonus target (20%)
- equity
- benefits
- career growth
- collaborative culture
- world-class team
Responsibilities- Build and ship classifiers and data pipelines for misbehavior detection
- Research cross-context monitoring systems and novel signal aggregation across sessions
- Explore monitoring signals from activations, actions, reasoning traces, and outputs
- Collaborate with infrastructure and data science teams to scale solutions
- Regularly share results with the Safety Oversight team
- Develop production-grade evaluation methods to measure safety and alignment
- Contribute to deployment-time safety surveillance and incident understanding
- Support collaboration with Gemini and GenMedia safety teams
Key requirements- PhD in Computer Science or related field, or equivalent practical experience
- 3 years of experience building and shipping technical products
- Experience in generative AI and Large Language Models (LLM) domain
- 3 years of experience developing code, running experiments, and analyses with coding agents
- Experience building large-scale, highly parallelised data pipelines
- Experience in data quality, automated evaluation design, and simple statistical modeling
- Ability to scale research solutions and work with production data
- Comfort using AI tools to push frontier of model capabilities
- collaborative
- analytical
- proactive
- large-scale data pipelines
- distributed systems
- ML safety and monitoring
