Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Member of technical staff at Anthropic and associate professor of computer science, data science, and linguistics at New York University on leave. His public homepage focuses on natural language processing, machine learning, and AI alignment.

Anthropic researcher on the Frontier Red Team focused on cyber misuse evaluation and threat modeling; previously a physics PhD student at UC Berkeley and now also mentors in the MATS program.

Co-founder and head of policy at Anthropic. He previously served as policy director at OpenAI, worked as a technology journalist, and writes the Import AI newsletter.

Associate Professor at the University of Toronto whose research spans deep learning, probabilistic modeling, and machine learning methods for science and AI safety.

Researcher focused on AI safety, reinforcement learning, and language models, with public work spanning red teaming, adversarial robustness, and model behavior.

Jesse Mu is a Research Scientist at Anthropic and a visiting researcher at Stanford University. His work spans machine learning, AI safety, reinforcement learning, and deep learning theory.

Associate Professor of Computer Science at the University of Toronto and director of the machine learning group, with research spanning probabilistic models and optimization algorithms.

Alignment researcher at OpenAI working on making AI understandable to and aligned with human values.

Anthropic co-founder and Chief Science Officer. Formerly a physicist at Johns Hopkins, he helped develop scaling laws for neural language models and works on the science and safety of large AI systems.

Anthropic researcher whose work includes reinforcement learning from human feedback and Constitutional AI; previously a Sherman Fairchild Postdoctoral Scholar in theoretical high-energy physics at Caltech.

Researcher at Anthropic working on alignment, reasoning, and evaluation for large language models.

Research scientist at Anthropic working on machine learning and AI safety.

Kshitij Sachan is a research scientist at Anthropic whose public homepage and Google Scholar profile highlight work on language models, reasoning, code generation, and machine learning systems.

Research scientist at Anthropic working on trustworthy AI and deceptive alignment.

AI safety researcher who led Anthropic's Safeguards Research Team and worked on jailbreak robustness, automated red teaming, and monitoring for misuse and misalignment.

Zachary Witten is a member of technical staff at Anthropic.

Research scientist at Anthropic focused on scalable oversight, AI safety, and language model evaluation; previously worked at New York University and Google.

Member of Technical Staff at Anthropic and cofounder of Oulipo Labs, working on language model safety, evaluations, and scientific forecasting.

Co-founder and head of alignment science at Anthropic.

Research scientist at Anthropic interested in understanding neural networks and applying that understanding to alignment.

Buck Shlegeris is a Member of Technical Staff at Anthropic whose public homepage focuses on AI safety, model evaluations, and alignment.

Member of Technical Staff at Anthropic and PhD student at Carnegie Mellon University focused on AI safety, evaluations, and oversight of large language models.

Member of technical staff at Anthropic working on alignment science and the evaluation of hidden objectives in language models.

Research scientist at Anthropic and former professor of theoretical astrophysics at Stony Brook University.