Research Engineer, Agentic Security for Gemini, DeepMind
Mountain View, CA, USA +1 · Full-time · Posted Oct 8, 2026
- Location
- Mountain View, CA, USA
San Francisco, CA, USA - Experience
- 1 year of experience in training or fine-tuning generative models to improve capabilities.
- Education
- Master's degree in Computer Science or related quantitative field.
About the job
Builds and evaluates training data, tools, and safeguards to improve Gemini’s security and privacy across agentic capabilities while preserving utility.
What you'll do
- Build post-training data and tools to improve Gemini’s security and privacy, and integrate improvements into Gemini.
- Coordinate with teams developing Gemini’s tool use, coding, and other agentic capabilities; improve adversarial evaluation and out-of-model guardrails.
- Develop reusable protection libraries and frameworks, and share knowledge through publications, open source, and education.
What they ask for
- Master’s degree in Computer Science or a related quantitative field.
- 1 year of experience training or fine-tuning generative models to improve capabilities.
- Experience in machine learning safety, security, privacy, or alignment.
- Experience with a machine learning framework such as JAX or PyTorch, and building readable, reusable ML software.
Nice to have
- Experience with JAX, PyTorch, or similar machine learning platforms.
- Python experience demonstrated through strong artifacts in readable, scalable, reusable ML software.
Pay
$174000 - $252000 (USD) (base salary)
Tools
Gemini, JAX, PyTorch, Python
AI focus
Improve Gemini’s security and privacy through post-training data and tools, adversarial evaluation, and safeguards for agentic capabilities.
Model-written summary · the original posting is authoritative