PatronusAI

Patronus AI is a frontier research lab developing simulation research and infrastructure to accelerate progress toward human-aligned artificial general intelligence. Its work centers on Digital World Models: systems that predict and simulate AI agents’ actions in digital workflows. These models can generate interactive digital worlds and high-value training simulations, helping frontier models learn from varied, realistic tasks and supporting continual learning.

The company’s simulation environments reflect human work across research, software development, customer service, product applications, and finance. They are designed to test capabilities such as deep research, multi-turn dialogue, long-horizon planning, and memory. Patronus AI offers products and resources including its Core Platform, Percival, reinforcement-learning environments, and tools for AI reliability and LLM testing.

Its research portfolio includes Lynx, a hallucination-detection model; FinanceBench, a benchmark for financial question answering; BLUR, a dataset for evaluating responses to tip-of-the-tongue queries; and GLIDER, an evaluation model focused on reasoning and explainability. Patronus AI also provides documentation, guides, research publications, and a playground for exploring its digital worlds.