PNNL

Ted Fujimoto is a Research Scientist specializing in AI safety, security, and foundational research at the intersection of machine learning, national security, and autonomous science. At Pacific Northwest National Laboratory (PNNL), he contributes to cutting-edge research in reinforcement learning, multi-agent systems, adversarial machine learning, generative AI, AI security, and open-ended AI.

Fujimoto plays a key role in the large language model reasoning component of the Theseus autonomous science project, which explores autonomous science through large language models. He is currently designing robust, cooperative multi-agent systems capable of operating in adversarial environments, with applications in autonomous cybersecurity. His work includes implementing scalable algorithms in PyTorch and leading the development of a time series foundation model integrated with submodular optimization for Earth systems science.

His research contributions include:

- Developing reliability protocols to quantify the impact of distribution shift on reinforcement learning performance.

- Assisting nuclear security experts with understanding AI and its impacts on arms control, national security, and diplomacy.

- Investigating emergent antagonistic behavior in reinforcement learning agents without access to opponent rewards—demonstrating novel attack strategies using entropy-based exploitation. This work earned the PNNL Postgraduate Research Symposium Laboratory Mission Award.

Beyond PNNL, Fujimoto was selected as one of three national AIxNuclear Fellows by the Berkeley Risk and Security Lab and the Council on Strategic Risks. In this role, he conducts open-source research on AI escalation risks, verification of dangerous capabilities, and AI’s role in nuclear stability.

Fujimoto holds a master’s in computer and information technology from the University of Pennsylvania and a bachelor’s in mathematics from the University of California, Berkeley.

Outside of research, he enjoys biking, reading philosophy, and exploring the nuances of coffee tasting.

• Causal analysis
• Computer vision
• Data science
• Deep learning
• Discrete mathematics
• Epistemology
• Game theory
• Graph analytics
• Machine learning
• Mathematical logic
• Multiagent systems
• Philosophy of artificial intelligence
• Python
• Reinforcement learning
• Risk analysis

• MS in computing science and computer and information science, University of Pennsylvania
• BA in math science, University of California, Berkeley

2024

• Marquez A., T.C. Fujimoto, and T.J. Stavenger. 2024. Assurance of Reasoning Enabled Systems (ARES). PNNL-36775. Richland, WA: Pacific Northwest National Laboratory. Assurance of Reasoning Enabled Systems (ARES)

2021

• Fujimoto T.C., T.J. Doster, A. Attarian, J.M. Brandenberger, and N.O. Hodas. 2021. "The Effect of Antagonistic Behavior in Reinforcement Learning." In AAAI-21 Workshop on Reinforcement Learning in Games, February 8, 2021, Virtual. Menlo Park, California:Association for the Advancement of Artificial Intelligence. PNNL-SA-157666.