We aim to turn production data into intelligence on the safety of deployed AI models. Safety Oversight is a new team tasked with using large-scale production traffic and a variety of automated evaluation methods to monitor the safety and alignment of deployed models. Our work will ensure we measure the real-world efficacy of our safety stack—both of in-model safety training and out-of-model safety mitigations to ensure we are effective in our goal of deploying safe models that are used for widespread public benefit.
The Safety Oversight team sits within the GenAI safety organization and is accountable for ensuring that when a model safety issue occurs in production, or when a user is misusing our model at scale, we detect and understand it, so the safety risk can be rapidly mitigated. We will collaborate closely with teams working on safety training and evaluation for Gemini and GenMedia models.
The Generative AI (GenA)I Safety team operates in a fast-paced, highly collaborative environment. We take the possibility of tangibly dangerous model capabilities seriously as AI advances, and we believe that proactive monitoring and deployment-time oversight are critical for safe AI development.
Artificial intelligence will be one of humanity’s most transformative inventions. At Google DeepMind, we are a pioneering AI lab with exceptional interdisciplinary teams focused on advancing AI development to solve complex global challenges and accelerate high-quality product innovation for billions of users. We use our technologies for widespread public benefit and scientific discovery, ensuring safety and ethics are always our highest priority.
We are pushing the boundaries across multiple domains. Our global teams offer diverse learning opportunities and varied career pathways for those driven to achieve exceptional results through collective effort.
Responsibilities
- Build classifiers and data pipelines to detect model misbehavior and misuse end-to-end.
- Research and develop cross-context monitoring systems to detect coordinated harms, developing novel signal aggregation methods across disparate user sessions to identify large-scale attack vectors.
- Think critically about novel methods for monitoring using model activations, actions, chains-of-thought and final answers.
- Collaborate closely with infrastructure teams and data scientists to scale your work and regularly share results with the wider safety team.
Qualifications
Minimum qualifications:
- PhD in Computer Science, a related field, or equivalent practical experience.
- Experience in the domain area of generative AI and Large Language Models (LLM).
- Experience building and shipping technical products.
Preferred qualifications:
- Master’s degree or PhD in Engineering, Computer Science, or a related technical field.
- 3 years of experience developing code, running experiments and analyses collaboratively with coding agents.
- Experience building highly parallelised data pipelines, working on data quality, automated evaluation design and simple statistical modeling.
- Proven ability in approaching new research questions and implementing technical solutions for them at scale.
- Ability to use AI every day to build and find ways to push the frontier of model capabilities to accelerate work.