Patronus AI, a San Francisco-based startup founded in 2023 by former Meta AI researchers Anand Kannappan and Rebecca Qian, has raised $50 million in a Series B round led by Greenfield Partners, with participation from Notable Capital, Lightspeed, Datadog, and Samsung. The raise brings total funding to $70 million and follows revenue growth of fifteen-fold over the past year.
The company builds simulated digital environments, which it calls digital world models, that replicate websites and internal systems to evaluate how AI agents perform across complex, unpredictable real-world scenarios. Agents are stress-tested using reinforcement learning, which rewards successful task completion and penalises errors, allowing model developers to identify shortcuts and failure modes before deployment. Kannappan compared the approach to how Waymoused synthetic environments to prepare autonomous vehicles for rare hazards.
Patronus argues that benchmark scores alone are insufficient evidence that an AI agent can reliably complete multi-step tasks such as financial analysis or travel booking. Its simulated environments currently focus on software engineering and finance, with expansion planned into harder-to-verify domains.
Glenn Solomon, managing director at Notable Capital, described demand from frontier AI labs and enterprise customers as nearly insatiable, noting that virtually every major lab has become a customer. Unlike human-data firms such as Mercor and Surge, Patronus conducts agent evaluation without human involvement, positioning itself primarily against the internal evaluation teams that AI labs have built themselves.
