A major public research university hosting several prominent AI safety research groups, including work on formal neural network verification, adversarial robustness, and AI agent security benchmarks.
A major public research university hosting several prominent AI safety research groups, including work on formal neural network verification, adversarial robustness, and AI agent security benchmarks.
People– no linked people
Updated 05/18/26Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 05/18/26The University of Illinois Urbana-Champaign, founded in 1867, is a flagship public research university with over 60,000 students and approximately 12,900 employees. Its Siebel School of Computing and Data Science is consistently ranked among the top public CS programs in the United States and is the home of several research groups working directly on AI safety problems. The FOrmally Certified Automation and Learning (FOCAL) Lab, led by Assistant Professor Gagandeep Singh, focuses on developing AI systems with formal mathematical guarantees about their behavior and safety. Research areas include neural network verification, certified adversarial robustness, formal methods for large language models, securing agentic AI, and neurosymbolic approaches. The lab is sponsored by NSF, Amazon, Google, Qualcomm, Bloomberg, and Open Philanthropy. The Secure Learning Lab (SL2), led by Abbasi Associate Professor Bo Li, investigates trustworthy machine learning with a focus on adversarial robustness, privacy, and generalization guarantees. Bo Li was selected as one of 27 projects in the Schmidt Sciences AI Safety Science Program (announced March 2025, $10M total), where her work involves building virtual red-teaming environments for systematically evaluating AI systems. She also co-founded Virtue AI, which raised $30M in seed and Series A funding in April 2025. Professor Daniel Kang's research group focuses on AI agent capabilities in cybersecurity contexts, developing benchmarks (including CVE-Bench, recognized with a SafeBench award and ICML spotlight) to measure whether LLM agents can autonomously exploit real-world software vulnerabilities. Kang was also selected for the Schmidt Sciences AI Safety Science Program and mentors at the MATS (ML Alignment Theory Scholars) program. Former UIUC PhD student Mantas Mazeika (graduated 2024) co-authored influential AI safety work including the HarmBench automated red-teaming framework and foundational x-risk analysis research, and is now a researcher at the Center for AI Safety. In 2025, UIUC and Capital One jointly launched the ASKS Center (Capital One Illinois Center for Generative AI Safety, Knowledge Systems, and Cybersecurity), with Capital One providing $3M over five years to support faculty research and PhD fellowships focused on generative AI safety, security, and trustworthiness.
Theory of Change
Updated 05/18/26UIUC's AI safety-relevant researchers pursue multiple complementary paths to reducing risk. The FOCAL Lab pursues formal verification approaches — proving mathematically that neural networks satisfy safety properties, which could enable reliable deployment in high-stakes settings. Bo Li's group works on adversarial robustness and red-teaming infrastructure so that dangerous capabilities in AI systems can be systematically identified and addressed before deployment. Daniel Kang's cybersecurity benchmarking work quantifies the offensive capabilities of AI agents, providing policymakers and labs with empirical evidence needed to make deployment and access decisions. Collectively, these efforts aim to produce tools, benchmarks, and theoretical frameworks that make it possible to evaluate and improve the safety of AI systems in a rigorous, reproducible way.
Grants Received
Updated 05/18/26Projects– no linked projects
Updated 05/18/26Discussion
No comments yet. Be the first to share your thoughts.