Research Scientist, AI Safety Alignment

Sourced from foundrole.com

Google DeepMind's Deep Understanding team seeks a Research Scientist to advance AI safety and alignment. You will conduct empirical research on machine behavior, evaluate and red-team large language models, and develop scalable safety mitigations for frontier systems. The role blends scientific inquiry with practical engineering to ensure models are safe and reliable in real-world use.

• Drive research projects by defining data structures, frameworks, and evaluation metrics.
• Identify gaps in existing technology and engage stakeholders to address them.
• Author research papers to share findings and impact the research community.
• Develop long-term research strategy and collaborate with internal and external partners.
• Evaluate, steer, and red-team LLMs and frontier systems to implement safeguards.

• PhD in Computer Science, Machine Learning, AI, Mathematics, Physics, or related STEM field, or equivalent experience.
• 1 year of experience in AI safety, alignment, or mechanistic interpretability research.
• Experience with LLMs or frontier systems, including evaluation, steering, red teaming, or safeguards.
• Proficiency in Python or C++ and deep learning frameworks (e.g., JAX, PyTorch).
• First-author publications at ML conferences (e.g., NeurIPS, ICML, ICLR, COLM).

• Post-doctoral or industry experience in prototype-driven safety projects.
• Interdisciplinary research bridging cognitive science, computer science, and ethics.
• Expertise in post-training alignment, reinforcement learning, and reasoning models.
• Track record of translating alignment findings into production-grade mitigations.
• Strong technical communication and stakeholder influence skills.