9/17/2026 Rudy San Miguel
Bo Li, CS and Abbasi Associate Professor at the Siebel School of Computing and Data Science in the Grainger College of Engineering, along with her Virtue AI team have been hired by Meta Superintelligence Labs (MSL) to bolster AI agent security. Founded in 2014, Virtue AI’s mission is a “unified platform ensuring the security, governance and compliance of every AI system.”
Written by Rudy San Miguel
Meta Superintelligence Labs (MSL) has hired Virtue AI founders and its team to bolster AI agent security. Virtue AI, a security startup, was founded by Bo Li, CS and Abbasi Associate Professor at the Siebel School of Computing and Data Science in the Grainger College of Engineering, along with Sanmi Kovejo, CS assistant professor at Stanford and adjunct associate professor at Siebel School, and Dawn Song, CS professor and director of RDI at UC Berkeley. Founded in 2014, Virtue AI’s mission is a “unified platform ensuring the security, governance and compliance of every AI system.” Meta’s hiring of its original founders and VIrtue AI team validates that vision.
For Li, this marks an exciting new chapter.
“When we started Virtue AI, we believed that security and trust would become one of the most important bottlenecks to deploying increasingly powerful AI systems in the real world,” Li said. “As AI evolves from models that primarily answer questions to agents that can reason, take actions and become deeply personalized, making these systems secure, reliable and trustworthy becomes even more critical.”
She says that working with Meta gives Virtue AI the opportunity to bring the technologies and research they have developed to a much larger scale. “We can work toward enabling secure and trustworthy personalized agents that have the potential to benefit billions of people,” Li said. “It is also incredibly exciting to see ideas that began in academic research and early prototypes become part of real-world systems at unprecedented scale.”
Li and her partners grew Virtue AI out of more than a decade of research in AI security, safety, privacy and trustworthy machine learning. They have also done a series of research for AI security and safety. Their paper on physical adversarial attacks has been cited over 4,000 times. Another paper on foundation model red teaming and risk assessment won best paper at NeurIPS and best scientific paper at the National Security Agency. In addition, Li's group has provided the first agent poisoning attack and analysis, as well as the first agent security protection and guardrail systems.
Li’s team has always been on top of rapidly-evolving AI and the inherent security risks. She says foundation models have become dramatically more capable.
“We saw a fundamental shift happening: AI was moving very quickly from research prototypes into systems that enterprises and individuals depend on every day,” she said. “At the same time, the security and safety infrastructure around these systems was not evolving nearly as quickly.”
Li says she and her partners founded Virtue AI to help close that gap.
“Our vision was to build a comprehensive trust and security layer for AI, from automated risk assessment and red teaming to runtime protection, guardrails, agent security, governance and compliance,” she said.
Li is certain of one thing after 12 years of groundbreaking work with her team: AI is only at the beginning of its transformation.
“The first wave of foundation models showed us how AI can understand and generate information. The next wave, AI agents, will increasingly be able to reason, plan, use tools, interact with other agents and take actions on behalf of people,” she said. “Eventually, highly personalized agents could understand our preferences and context, help us with everything from everyday tasks to highly complex professional work and recursively improve themselves.”
Li and her team know that the opportunities are endless, but so are the risks, making trust and security foundational.
“The more capable and autonomous an AI system becomes, the more important it is that we understand what it is doing, protect it from manipulation, preserve users' privacy and ensure that its actions remain aligned with users' intentions,” she said.
Because of this, Li believes the most important AI advances will come not only from making models more intelligent, but making intelligent systems more secure, reliable and trustworthy. “If we can get both sides right,” Li says, “AI has the potential to become an extraordinarily positive technology for society.”
From working with leading enterprises and AI organizations, she says those collaborations gave them a lesson in what it takes to secure AI in real-world environments. She also expressed gratitude to her team, customers, partners, investors and research collaborators who shared Virtue AI’s mission and helped get them where they are today.
“We started with a research-driven belief that AI security would become fundamental to the future of AI, and we were fortunate to work with an exceptional team and with customers and partners who shared that conviction,” Li said.
She added, “What excites me most is that the mission continues, and at a much larger scale. There is still an enormous amount to learn and build. As both a researcher and an entrepreneur, I have always cared deeply about turning fundamental research into technologies that can have meaningful real-world impact. I am excited to continue learning, continue pushing the frontier of AI security and trustworthy AI, and ultimately help build AI systems that people around the world can confidently rely on.”
Grainger College affiliations:
Bo Li is an Illinois Grainger professor of computer science in the Siebel School of Computing and Data Science.
Sanmi Koyejo is an Illinois Grainger adjunct associate professor of computer science in the Siebel School of Computing and Data Science.