Patronus AI has raised $50M in a Series B led by Greenfield Partners, bringing total funding to $70M. The company builds simulated digital environments called Digital World Models to stress-test AI agents before production deployment. Borrowing from Waymo's synthetic training approach, Patronus uses reinforcement learning inside replicas of websites and internal systems to catch agent shortcuts and failures. Revenue has grown fifteenfold in the past year, with virtually every frontier AI lab as a customer. Current simulated environments cover software engineering and finance, with plans to expand into harder-to-verify long-horizon tasks. The company positions itself against internal AI lab evaluation teams and human annotation firms, arguing its automated, scalable approach is superior.
