Patronus AI raises $50 million to stress-test AI agents

On Thursday, Patronus AI announced its latest funding milestone to expand its platform for evaluating artificial intelligence. The San Francisco-based startup, which was founded in 2023 by Anand Kannappan and Rebecca Qian, builds simulated digital environments to stress-test AI agents. These agents are software systems that are evolving to autonomously execute multi-step, complex tasks, addressing the need for reliability in complex, real-world jobs.

The company’s funding breakdown includes:

• Series B funding: $50 million, led by Greenfield Partners.
• Total funding to date: $70 million.

Other investors participating in the round include Notable Capital, Lightspeed, Datadog, and Samsung.

The investment follows a period of commercial growth, with Patronus AI’s revenue growing 15-fold over the past year. The startup uses what it calls “digital world models”—simulated environments used to create replicas of websites and internal systems—to evaluate how agents perform. Within these simulated environments, agents are stress-tested after training using reinforcement learning, which is a training method that iteratively rewards successful task completion and penalizes errors.

While the startup currently focuses on verifiable problems that can be immediately checked, Kannappan explained that there are many other areas that are very hard to verify. The ultimate goal is to support long-running autonomous operations. “We want to be able to actually create the environment in which you can operate an agent that can run for 10 hours or 10 days or 10 weeks,” said Kannappan.

According to Glenn Solomon, a managing director at Notable Capital, demand for the company’s simulated environments is nearly insatiable. Solomon noted that the startup is effective at spotting shortcuts and holding models accountable. Rather than competing directly with human-data firms like Mercor or Surge, Patronus AI primarily competes against the internal teams that AI labs build to evaluate agent behavior. By automating the evaluation process without human involvement, the startup aims to identify where agents fail to complete tasks correctly and ensure they perform reliably.

Why it matters

As AI agents evolve from simple chatbots to autonomous workers, the ability to verify their performance in complex, real-world scenarios is becoming a critical bottleneck for enterprise adoption.