Senior Research Scientist, AI Deception

Job Summary

To advance AI safety research, the full-time Senior Research Scientist, AI Deception will lead projects focused on deception detection, model monitoring, and training methodologies, working remotely or in-person in Berkeley.

Key responsibilities

• Lead research projects on white-box deception detection and probe-based monitoring, defining research questions and assessing results
• Run experiments on large models, diagnosing training issues and designing stress-test scenarios
• Mentor Technical Staff and collaborate with infrastructure teams to enable method adoption by frontier labs

Required qualifications

• 3+ years of experience in post-training or related efforts in a frontier AI environment
• Hands-on experience with large-scale model training, ideally with models of 400B+ parameters
• Strong proficiency in Python and PyTorch, with experience in shared research codebases
• Demonstrated research judgment for experimental design and result evaluation
• Willingness to work in U.S. time zones, with a preference for in-person presence in Berkeley