Patronus AI
Patronus AI is a San Francisco-based AI evaluation and security company founded in 2023 by Rebecca Qian and Anand Kannappan, both formerly at Meta.
Profile
Patronus AI sells an automated platform that helps enterprises test, evaluate, and monitor large language models for accuracy, safety, and compliance before and after deployment.
Patronus AI is a San Francisco-based AI evaluation and security company founded in 2023 by Rebecca Qian and Anand Kannappan, both formerly at Meta. The company provides an automated platform for enterprise development teams to test, evaluate, and monitor large language models (LLMs) at scale. Its product suite includes LLM performance scoring, adversarial test case generation, hallucination detection, benchmarking, and real-time agent monitoring via its Percival platform.
The company has raised a total of $20 million in disclosed funding: a $3 million seed round led by Lightspeed Venture Partners in September 2023, and a $17 million Series A led by Notable Capital (formerly GGV Capital) in May 2024, with participation from Lightspeed, Datadog, and others. As of 2025, PitchBook reports 34 employees and a private, venture-backed status. ZoomInfo estimates revenue under $5 million.
Patronus AI has published several open-source research benchmarks and models, including FinanceBench (a financial Q&A benchmark), Lynx (a hallucination detection model), and BLUR (an agent evaluation dataset). The company serves Fortune 500 enterprises and leading AI companies, with a disclosed partnership with CARIAD, Volkswagen's software division, announced in June 2025. The company's stated vision is 'scalable oversight'—using AI to evaluate AI, with humans as overseers.
In early 2025, Patronus AI launched a multimodal LLM-as-a-judge for image evaluation. The company faces a competitive landscape with at least 63 competitors, including Clarifai, and operates in the rapidly evolving LLM evaluation and guardrails market.
Who buys this
- Fortune 500 enterprises deploying generative AI applications
- Leading AI companies needing model evaluation and guardrails
- Financial services firms requiring domain-specific LLM benchmarks (e.g., FinanceBench)
- Automotive and software companies integrating LLMs into products (e.g., CARIAD/Volkswagen)
- Enterprise development teams seeking automated, scalable LLM testing and monitoring
Publicly disclosed clients
- CARIAD (Volkswagen's software company)
Strengths and what to watch
Strengths
- Published multiple open-source, industry-first benchmarks (FinanceBench, Lynx, BLUR) that demonstrate research credibility and attract developer adoption.
- Backed by notable investors including Notable Capital, Lightspeed Venture Partners, and Datadog, providing both capital and strategic distribution channels.
- Product is API-first and model-agnostic, allowing integration with any LLM and supporting a broad range of evaluation dimensions (hallucination, PII, copyright, brand alignment).
Watch for
- Revenue is undisclosed and likely still sub-$5 million (per ZoomInfo), with only 34 employees (per PitchBook), raising questions about unit economics and path to scale in a competitive market with 63+ rivals.
- The company's website underwent a significant rebranding or content shift in early 2025, now describing itself as a 'frontier lab training Digital World Models'—a pivot from its original LLM evaluation focus that may confuse existing customers or signal strategic uncertainty.
- No disclosed customer count or churn data; the only named client is CARIAD (Volkswagen), and it is unclear how many Fortune 500 enterprises are paying customers versus pilot users.
Recent moves
Key Information
- Industry
- AI Observability
- Founded
- 2023
- Headquarters
- San Francisco
Frequently Asked Questions
What does Patronus AI do?
Patronus AI provides an automated platform for enterprises to test, evaluate, and monitor large language models for accuracy, safety, and compliance. It offers hallucination detection, adversarial test generation, and real-time agent monitoring through its Percival platform.
Who founded Patronus AI and when?
Patronus AI was founded in 2023 by Rebecca Qian and Anand Kannappan, both former employees at Meta. The company is based in San Francisco and focuses on AI evaluation and security for enterprise LLM deployments.
How much funding has Patronus AI raised?
Patronus AI has raised a total of $20 million in disclosed funding. This includes a $3 million seed round led by Lightspeed Venture Partners in September 2023 and a $17 million Series A led by Notable Capital in May 2024.
What open-source benchmarks has Patronus AI released?
Patronus AI has published several open-source benchmarks and models, including FinanceBench for financial Q&A, Lynx for hallucination detection, and BLUR for agent evaluation. These tools help developers test LLM performance in specific domains.
Who are Patronus AI's main customers?
Patronus AI serves Fortune 500 enterprises and leading AI companies. Its disclosed client includes CARIAD, Volkswagen's software division, and it targets financial services firms and enterprise development teams needing automated LLM testing.
What are Patronus AI's main competitors?
Patronus AI faces competition from at least 63 companies in the LLM evaluation and guardrails market, including Clarifai. The company operates in a rapidly evolving space with rivals offering similar testing and monitoring solutions.
Sources
- www.patronus.ai — Company description, product offerings, research publications (Lynx, FinanceBench, BLUR, GLIDER), and vision statement about Digital World Models.
- pitchbook.com — Funding history ($17M Series A in May 2024, $3M seed in Sep 2023), employee count (34), status (private, venture-backed), and industry classification.
- www.patronus.ai — Details of Series A round ($17M, led by Notable Capital, with Lightspeed and Datadog), customer base (Fortune 500), and product capabilities.
- www.patronus.ai — Seed round announcement ($3M, led by Lightspeed), founding team background (Rebecca Qian, Anand Kannappan, Meta), and initial product vision.
- www.zoominfo.com — Revenue estimate (<$5M), employee count (1-10, though PitchBook shows 34), and company description.
- patronus.ai — List of product launches, partnerships (CARIAD/Volkswagen), and research releases (multimodal LLM-as-a-judge, Lynx, FinanceBench).