Gray Swan

Not a pentest with an AI wrapper. Gray Swan's red teaming service is powered by Arena threat intelligence and delivered by the team that breaks frontier models pre-release.

Find AI edge cases, policy gaps, and failure modes before production. Gray Swan's automated red teaming surfaces what internal testing can't.

Runtime governance for AI agents. Gray Swan enforces behavioral policies on every action, every interaction.

Gray Swan prevents AI agents from exposing customer PII, proprietary data, or sensitive information by enforcing data policies at runtime.

The page lists seven warning signs of misleading AI vendor claims, including overstated framework coverage.

Gray Swan introduces AgentHarm, a benchmark measuring harmfulness in LLM agents, and finds limited robustness in models like GPT-4o and Claude Sonnet 3.5.

A new study from Stanford and Gray Swan finds that purpose-built AI agents can outperform most human cybersecurity professionals in real-world penetration testing—at a fraction of the cost.. Updated: Dec 11, 2025

Google DeepMind and Anthropic sponsor the UK AISI Agent Red-Teaming Challenge, increasing the prize pool to $170k.

Gray Swan is the safety and security provider for the AI era.

Shade identifies vulnerabilities in AI evaluations that automated tests may miss; Gray Swan provides a demo of its autonomous red-teaming platform.

Gray Swan offers personalized demos to help organizations deploy AI securely.

Gray Swan provides enterprise-grade security solutions for LLMs, developed by the pioneers of AI vulnerability research. Deploy AI with confidence using just two lines of code.

Gray Swan is the safety and security provider for the AI era. This is Gray Swan's policy on product's acceptable use.