Gray Swan, an AI security company that originated from Carnegie Mellon University’s AI safety research programme and is embedded in the safety evaluation processes of the world’s leading frontier labs, has raised $40m in a Series A funding round.
The round was co-led by Wing Venture Capital and Madrona, with participation from Obvious Ventures, Snowflake Ventures, Hudson River Trading, Samsung Next, and existing backer Magarac Venture Partners. The capital will be used to accelerate Gray Swan’s commercial operations, strengthen its relationships with frontier AI labs, and grow the team supporting enterprise AI deployments.
Gray Swan is named as a security evaluator in 11 recent frontier model system cards — including those published by Anthropic, OpenAI, and Meta — with its benchmarks woven into the pre-release safety processes for some of the most advanced AI models in existence. The company says no other AI security firm operates with this level of simultaneous access and trust across multiple leading frontier labs.
The company’s co-founders, Matt Fredrikson and Zico Kolter, spent more than a decade researching how AI systems behave under adversarial conditions before founding Gray Swan. Their academic work preceded much of the industry’s awareness of threats such as prompt injection, jailbreaks, and data exfiltration — risks that have shifted from theoretical to operational as AI agents gain access to enterprise data and begin executing autonomous workflows. Gray Swan was created to translate that research into practical, market-ready security solutions.
The firm currently serves more than 20 customers across frontier labs and large enterprises. It has also formed a partnership with Snowflake, through which Gray Swan’s runtime AI protection is natively integrated into Snowflake’s AI ecosystem, allowing enterprises to embed security capabilities directly into the platform where they build and scale AI applications.
Gray Swan’s platform is built around three components. Cygnal delivers real-time AI monitoring and protection, enforcing custom policies that organisations define based on their specific risk profiles. Shade is an automated red-teaming agent that runs adversarial tests against AI models and agents as part of pre-deployment testing and continuous integration pipelines, surfacing vulnerabilities before they reach production. Arena is a global competition network of more than 15,000 researchers and security professionals who actively probe AI models for weaknesses. Together, Arena has generated more than one million real-world attack trajectories, which are used to train and update the models underpinning both Cygnal and Shade.
Gray Swan co-founder and chief scientist Zico Kolter said, “AI applications are growing at an unprecedented rate, and at Gray Swan we want to ensure that these deployments can continue without sacrificing reliability and security. We have spent our careers conducting research into the safety of AI systems, and we are putting this into practice to ensure the companies can deploy AI with confidence.”
Gray Swan CEO and co-founder Matt Fredrikson said, “AI is moving faster than any technology we’ve seen, and security hasn’t kept pace. Our mission is to empower the world to use AI safely and securely, and this funding lets us pursue the mission on pace with the frontier. Our partnership with Snowflake is a perfect example: bringing the same security trusted by frontier labs to the platform where enterprises are already building on their data.”
Copyright © 2026 RegTech Analyst
Copyright © 2018 RegTech Analyst
