Patronus AI builds tools that test and train AI systems before they are trusted with real work. The company announced today that it raised $50 million in a Series B round led by Greenfield Partners. Existing investors Lightspeed Venture Partners, Notable Capital, Datadog, and Samsung joined the round, along with Factorial Capital, Gokul Rajaram, and other AI and software executives. The new funding brings Patronus AI's total raised to $70 million.
The company was founded by Anand Kannappan and Rebecca Qian, two former Meta AI researchers. Their idea is simple to state but hard to pull off: AI labs need better ways to check whether their systems can actually be trusted, especially as those systems take on longer and more complicated jobs.
That need is becoming urgent. A 2026 industry report from Digital Applied found that 88% of AI agent pilots never reach production. Enterprise leaders most often point to gaps in evaluation and testing as the reason. The problem holding back AI agents is rarely about building them. It's about proving they work reliably once they're out in the real world.
For the first few years of the generative AI boom, companies judged AI models using fixed tests called benchmarks. A model answers a set list of questions and gets scored, similar to a school exam. That worked fine when AI was mostly used to answer questions or write short pieces of text.
It works much less well now. AI is increasingly asked to act as an "agent," meaning it carries out long, multi-step jobs on its own, like resolving a customer complaint from start to finish or fixing a bug in a company's software. A fixed test cannot capture everything that might go wrong, or right, across a long task with many possible paths.
An agent has to navigate unfamiliar software, juggle several steps at once, and sometimes recover after making a mistake. That kind of work needs far more varied practice than a static quiz can offer. This is the gap Patronus AI says it is trying to close with what it calls "Digital World Models," a system designed to generate large numbers of realistic digital scenarios where AI agents can train and be evaluated, instead of relying on one fixed set of test questions.
Greenfield Partners, the lead investor in this round, described the bet as one on infrastructure rather than on any single product.
"Patronus AI is tackling one of the most important infrastructure problems in artificial intelligence. The future of AI will depend on systems that can learn and operate reliably in complex environments, and simulations are becoming essential to making that possible." — Itay Inbar, Partner at Greenfield Partners
With the new capital, Patronus AI plans to grow its research team, expand its engineering staff, and invest in the computing power needed to build and run Digital World Models at scale. Training these simulated environments, and running them for many AI agents at once, takes a lot of computing power. A large share of the new funding will likely go toward that infrastructure.
Patronus AI says it also plans to speed up its sales and growth efforts. The company says it already works with most of the world's leading AI labs and major cloud computing providers, sometimes called hyperscalers. It also says its revenue has grown more than 15 times over the past year. That kind of growth suggests demand for testing and simulation tools is rising fast, as more companies move from experimenting with AI agents to actually putting them to work.
CEO and co-founder Anand Kannappan framed the company's goal around a problem that will only grow more pressing as AI systems take on more responsibility without a person checking every step.
"Manual review does not scale once AI systems begin operating across millions of workflows and decisions. That is why simulations matter. They create environments where AI systems can be tested, improved, and supervised before failures happen in production." — Anand Kannappan, CEO and Co-founder of Patronus AI
Patronus AI was founded just under three years ago. Anand Kannappan serves as CEO, and Rebecca Qian serves as Chief Technology Officer. The company identifies both as former Meta AI researchers. More broadly, Patronus AI says its team's experience spans organizations including Meta AI, Amazon AGI, and Google, covering AI evaluation, AI alignment, fairness, and embodied agents, which is the term for AI systems that interact with a physical or simulated environment.
In plain terms, Patronus AI builds practice environments for AI agents. Instead of grading a system on a fixed quiz, the company creates simulated versions of real digital tasks. That might mean using business software, researching across many documents, or troubleshooting a technical problem, while an AI agent works through the steps. Its newest technology, called Digital World Models, generates large numbers of these training scenarios automatically rather than building each one by hand.
The company has compared this approach to how Waymo uses world models to simulate road conditions and unusual situations that self-driving cars haven't encountered yet, rather than testing every single real road. Patronus AI says it's doing something similar, but for digital workflows instead of physical driving.
What sets the company apart, based on its own public statements, is that it positions itself as both a training ground and a testing ground: a place where AI systems can practice, and also a place where they get checked for reliability before being trusted with real tasks. As Kannappan put it in the company's announcement, static evaluations cannot show whether an agent can handle ambiguity or recover from failure over a long task.
Patronus AI's $50 million Series B was led by Greenfield Partners, a fund that focuses on backing technology companies at the early growth stage. The round also included existing backers Lightspeed Venture Partners, Notable Capital, Datadog, and Samsung, along with Factorial Capital, Gokul Rajaram, and other AI and software industry executives.
