Dan Hendrycks

PersonResearcher
Suggest Edit
Dan Hendrycks
RoleCAIS Director
Known ForAI Safety Benchmarks
AffiliationUC Berkeley
EducationPhD, UC Berkeley

Dan Hendrycks is an AI safety researcher and director of the Center for AI Safety (CAIS). He is known for creating influential benchmarks for measuring AI capabilities and safety, including MMLU (Massive Multitask Language Understanding) and various robustness evaluations.

Career

Academic Research

Hendrycks completed his PhD at UC Berkeley, where he developed numerous benchmark datasets that have become standard tools for evaluating language models. His work on measuring AI capabilities has influenced how the field tracks progress toward more powerful systems.

Center for AI Safety (2022-present)

As director of CAIS, Hendrycks has led efforts to advance AI safety research and advocate for responsible AI development. CAIS played a key role in organizing the 2023 "Statement on AI Risk" signed by leading AI researchers and executives.

Key Contributions

  • MMLU Benchmark: Created the Massive Multitask Language Understanding benchmark, widely used to evaluate language model knowledge and reasoning
  • Robustness Research: Developed methods for testing AI system robustness to distribution shift and adversarial inputs
  • AI Risk Advocacy: Organized the influential "Statement on AI Risk" comparing AI extinction risk to pandemics and nuclear war
  • Safety Benchmarks: Created evaluation suites for measuring dangerous AI capabilities

Research Interests

  • AI safety benchmarks and evaluations
  • Distributional shift and robustness
  • Measuring dangerous AI capabilities
  • AI governance and policy

Notable Works

  • "Measuring Massive Multitask Language Understanding" (2020) - MMLU benchmark paper
  • "Natural Adversarial Examples" (2021)
  • "Unsolved Problems in ML Safety" (2022)
  • "An Overview of Catastrophic AI Risks" (2023)

See Also

Last updated: December 7, 2025