Virtue AI

Virtue AI is built on 𝐨𝐯𝐞𝐫 𝐚 𝐝𝐞𝐜𝐚𝐝𝐞 of foundational research in adversarial AI: the kind that gets published at NeurIPS, recognized by the National Security Agency, and relied upon by Fortune 500 companies. 20+ PhDs. 10+ patents. 15+ papers at NeurIPS 2025 alone. Research on how agents get compromised, how memory and knowledge bases get poisoned, how code generation systems fail under adversarial conditions, and what it takes to defend enterprise AI at scale. See what's under the hood at virtueai.com/research

We are very excited to host the AI security Conference from Virtue AI! As AI systems and autonomous agents evolve at an unprecedented pace, security, safety, and trustworthiness have become the key barriers to real-world enterprise deployment. Building powerful AI is no longer enough, and we need to ensure these systems are secure, reliable, and aligned, which is now a fundamental challenge for the entire industry. At this conference, we are bringing together leaders across AI and AI security to discuss the future of AI, emerging risks in agentic systems, and the first principles needed to build secure and trustworthy AI. Looking forward to the discussions and to building the AI security community together. Come join us!

NEW RESEARCH: If your stack includes code generation in CI/CD or agent workflows, the most likely failure point isn't a dramatic breach; it's a model that executes a biased instruction... That's the gap BlueCodeAgent closes.

Welcome to our summer interns! Virtue AI was built on foundational research into how AI systems get attacked, compromised, and manipulated. That work doesn't stop here. We started our intern program because the next generation of researchers and builders should be doing this work with us, not after us. Excited to have this group in the building.

"Policies do not enforce themselves. You have to have good enforcement mechanisms. What your policies can do is help with the human readiness side. But ultimately, you do need enforcement for it to really work in the long run." - Christopher Campbell, MBA, CISSP (Director of AI Governance, Lenovo)   ICYMI: Last week, Virtue AI Co-Founder Sanmi Koyejo sat down with Campbell to talk through what enterprise AI governance actually looks like in practice. The gap (between what a policy says and what actually runs at the model and agent layer) is where enterprise AI risk lives right now. Virtue AI PolicyGuard closes that gap. Watch the conversation here: https://lnkd.in/gkX2f7Un

Too often, research insights stay in papers and operational lessons stay inside companies. CTRL+AI brings everyone into the same room, to elevate AI security together. June 4. San Francisco. Register now: https://lnkd.in/gDKH2ugQ Speakers include Sandeep Pombra Ph.D. (NVIDIA), Leon Bian (Capital One), Ravi Krishnamurthy (ServiceNow), Heather Ceylan (Box), Dawn Song (University of California, Berkeley, Virtue AI), Bo Li (University of Illinois Urbana-Champaign, Virtue AI), Sunil Agrawal (Glean), Sanmi Koyejo (Stanford University, Virtue AI), Carlos Guestrin (Stanford University, Virtue AI), Seth Spiel (Splunk), Wenbo Guo (UC Santa Barbara, Virtue AI), and Minzhou Pan (Virtue AI)

Earlier this year, the National Institute of Standards and Technology (NIST), issued a formal Request for Information (RFI) on the security of AI agent systems. Virtue AI was there to answer the call 📣 Read more👉 https://lnkd.in/ghjG_m3x

NVIDIA... Stanford University... ServiceNow... University of California, Berkeley... Box... Glean... Splunk... Capital One... University of Illinois Urbana-Champaign... AI security's top researchers and practitioners are assembling on June 4 in San Francisco. Join us! 👉 https://lnkd.in/gDKH2ugQ Enterprise AI security has a translation problem. Researchers are uncovering new attack surfaces, model vulnerabilities, and agentic failure modes. Security leaders are trying to ship AI safely under real constraints: compliance, business pressure, data risk, and production complexity. Too often, the two conversations happen separately. CTRL+AI is designed to close that gap.

𝐓𝐎𝐌𝐎𝐑𝐑𝐎𝐖! Join Sanmi Koyejo (Co-founder, Virtue AI / Associate Professor, Stanford University) and Christopher Campbell, MBA, CISSP (Director of AI Governance, Lenovo) for a discussion on what practical AI policy enforcement actually looks like in 2026 and beyond. Register now👉 https://lnkd.in/gSYGktcT   Make sure to stay until the end for a special Q&A. You don't want to miss it!   𝐓𝐡𝐞 𝐀𝐈 𝐏𝐨𝐥𝐢𝐜𝐲 𝐆𝐚𝐩: 𝐇𝐨𝐰 𝐑𝐮𝐥𝐞𝐬 𝐁𝐞𝐜𝐨𝐦𝐞 𝐄𝐧𝐟𝐨𝐫𝐜𝐞𝐦𝐞𝐧𝐭 📅 Thursday, May 14 ⏰ 10 a.m. PT

Using DTap, Virtue AI conducted large-scale evaluations of modern AI agents across realistic environments. The result? 𝐓𝐨𝐝𝐚𝐲’𝐬 𝐚𝐠𝐞𝐧𝐭𝐬 𝐚𝐫𝐞 𝐬𝐲𝐬𝐭𝐞𝐦𝐚𝐭𝐢𝐜𝐚𝐥𝐥𝐲 𝐯𝐮𝐥𝐧𝐞𝐫𝐚𝐛𝐥𝐞. 🫢