Gray Swan AI

Gray Swan AI was founded by the AI security researchers from Carnegie Mellon University who pioneered and published the foundational research on prompt injection and adversarial attacks against language models. We didn’t read about these attacks in papers; we wrote the papers.

We began by red teaming for frontier labs like OpenAI, Anthropic, Google DeepMind, Meta, and Amazon, which led us to build The Arena; now the world’s largest AI red-teaming platform with more than 12,000 active red teamers continuously uncovering new attack vectors. This constant stream of real-world adversarial behavior gives us threat intelligence that simply doesn’t exist anywhere else.

At AI Summit NY, we’re highlighting how these real attacks and real learnings directly inform the products we build. Our platform enables enterprise teams to battle-test AI systems in live environments, validate defenses with production-level realism, and gain the confidence they need to deploy safely at scale. With Gray Swan, organizations don’t just model theoretical threats—they train against the ones attackers are using today.