Patronus AI has raised a $50 million Series B led by Greenfield Partners, with participation from Lightspeed, Notable Capital, Datadog and Samsung, lifting its total funding to $70 million. The company, founded in 2023 by former Meta AI researchers Anand Kannappan and Rebecca Qian, builds simulated digital environments to evaluate AI agents before they are turned loose on real systems.
The product is essentially a testing ground: Patronus constructs 'digital world models' that replicate websites and internal enterprise systems, then runs AI agents through them to see where they take shortcuts, hallucinate, or fail to complete tasks correctly -- analogous to how autonomous-vehicle companies validate self-driving systems in virtual scenarios before public roads. As agents move from demos into production, this kind of pre-deployment stress-testing becomes essential infrastructure.
“As agents move from demos into production, this kind of pre-deployment stress-testing becomes essential infrastructure.”
The traction validates the timing. Patronus says it grew revenue 15-fold over the past year, with one investor describing demand as 'nearly insatiable,' and counts nearly all frontier AI labs plus many emerging startups as customers. That customer base -- the very labs building the most advanced agents -- is a strong signal: the people who understand agent failure modes best are paying Patronus to catch them.
The round fits the broader 2026 pattern of capital flowing to applied AI with measurable enterprise value, and specifically to the 'agent reliability' layer that has emerged as agents proliferate. Patronus competes with evaluation and observability players like LangSmith, Braintrust and a wave of AI-testing startups, as well as labs' internal tooling. Its current focus spans software engineering and finance, with plans to expand into harder-to-verify domains.
The bear case: evaluation tooling can be replicated, frontier labs may build it in-house, and the category is young and crowded. What to watch: whether Patronus can stay ahead as agent capabilities evolve, how it expands beyond engineering and finance, and whether agent evaluation consolidates into a standalone market or gets absorbed into broader AI-observability platforms.