The academics who studied AI safety for decades decided to ship it. One security solution for your entire AI stack.
The wordmark of a company whose entire job is keeping AI well-behaved. Virtue, as in good conduct - applied to machines that increasingly act on their own.
Here is a thing that is true about enterprise software in 2026: a lot of companies have decided they need artificial intelligence, and a somewhat smaller number of them have figured out what happens when that artificial intelligence does something they did not ask it to do. This is the gap Virtue AI is built to fill, and it is a real gap, because the tools that secure ordinary software were designed for software that does exactly what it is told. AI is not like that. A model can be talked into ignoring its instructions, fed poisoned training data, or tricked into confidently making things up - and none of those failure modes look like a buffer overflow.
Virtue AI was founded in 2024 by four people who have spent an unusual amount of their professional lives worrying about exactly this. Bo Li, the CEO, along with Dawn Song, Carlos Guestrin, and Sanmi Koyejo, are academics from Berkeley, Stanford, and Chicago whose research careers ran straight through adversarial machine learning and AI safety. The pitch is not subtle: these are the people who wrote some of the papers, and now they would like to sell you the product.
That is a more interesting business proposition than it sounds. AI safety spent years as an academic field and a philosophical debate, the kind of thing that produced conference talks rather than invoices. The bet Virtue AI is making is that it has quietly become a line item - that enterprises deploying AI agents will pay for the same rigor that used to live in a research lab. In April 2025 the company came out of stealth with $30 million in combined seed and Series A funding, co-led by Lightspeed Venture Partners and Walden Catalyst Ventures, with Prosperity7, Factory, Osage University Partners, and angels including Intel CEO Lip-Bu Tan. That is a lot of money for a field that, not long ago, mostly generated citations.
What the money buys is a platform that does three things. It attacks your AI on purpose, it stands guard in front of your AI in real time, and - as of early 2026 - it does both for the newest and twitchiest category of all: AI agents, the systems that do not just answer questions but take actions on your behalf.
The reason that last category matters is worth stating plainly. A chatbot that gets jailbroken says something embarrassing. An agent that gets jailbroken can call a tool, move data, or execute a transaction. The blast radius is different, and Virtue AI's newest product line is essentially an argument that the industry has not yet reckoned with it.
"Our team has dedicated decades to solving these exact problems. Virtue AI transforms that expertise into practical solutions."
Automated red-teaming that throws 100+ proprietary attack algorithms, 600+ attack vectors, and 320+ risk categories at your models and agents - jailbreaks, prompt injection, privacy leaks, hallucinations - before they ship.
Real-time guardrail models across text, image, audio, video, and code in 90+ languages. The company claims sub-10ms latency, up to 30x faster and 40-50% higher performing than alternatives. Includes PolicyGuard and per-modality guards.
An end-to-end platform for enterprise AI agents. AgentSuite-Red runs autonomous red-teaming in sandboxes; AgentSuite-Blue adds runtime MCP Guard, Action Guard, and Shadow AI discovery, plus access control and audit trails.
A security agent that reads internal company policies and external regulatory requirements, then automates the tedious compliance mapping that would otherwise be done by hand.
"Traditional computing security is insufficient for artificial intelligence use cases."
AI-security researcher who has become one of the field's most cited voices on adversarial ML and trustworthy AI. Leads Virtue AI's push to turn research into an enterprise platform.
Berkeley professor and one of the most recognized names in computer security and AI safety research.
Stanford professor and veteran machine-learning researcher with deep roots in applied ML systems.
Researcher focused on trustworthy and reliable machine learning, rounding out the founding quartet.
| Round | Amount | Date |
|---|---|---|
| Seed + Series A | $30,000,000 | Apr 2025 |
Co-led by Lightspeed Venture Partners and Walden Catalyst Ventures, joined by Prosperity7 Ventures, Factory, Osage University Partners, and angels including Intel CEO Lip-Bu Tan, Amarjit Gill, and Stanford's Chris Ré. The latest slice - a $22.5M Series A - closed around April 2025.
Virtue AI works with enterprises in finance, healthcare, and IT, plus frontier AI labs. Publicly referenced customers and partners include OpenAI, NVIDIA, Microsoft, Zoom, Uber, and Glean. Uber uses it for content-safety guardrails; Glean collaborates on emerging threats. The notable part: some of the labs building AI are also buying Virtue AI to secure it.
Bo Li, Dawn Song, Carlos Guestrin, and Sanmi Koyejo start the company to close the enterprise AI security gap.
Combined seed and Series A funding co-led by Lightspeed and Walden Catalyst; unveils VirtueRed, VirtueGuard, and VirtueAgent.
An AI-native security and governance platform for enterprise AI agents, spanning red-teaming, runtime protection, and access control.
It provides an enterprise AI security platform that red-teams AI models and agents, applies real-time multimodal guardrails, and enforces compliance so companies can deploy generative AI and AI agents safely.
It was founded in 2024 by AI-safety researchers Bo Li (CEO), Dawn Song, Carlos Guestrin, and Sanmi Koyejo, from Berkeley, Stanford, and Chicago.
The company announced $30 million in combined seed and Series A funding in April 2025, co-led by Lightspeed Venture Partners and Walden Catalyst Ventures.
VirtueRed (automated red-teaming), VirtueGuard (real-time multimodal guardrails), VirtueAgent (compliance agent), and AgentSuite (end-to-end agent security).
Frontier AI labs and enterprises in finance, healthcare, and IT, with publicly referenced customers including OpenAI, NVIDIA, Microsoft, Zoom, Uber, and Glean.
Sources: BusinessWire, SecurityWeek, Axios, TechStartups, PR Newswire, Crunchbase, PitchBook, virtueai.com. Metrics are company-reported and approximate.