AI Security
Oct 20, 2026
Gray Swan relaunches AI red-teaming Arena with new challenges and increased prize money
Oct 20, 2026
AI Summary
Gray Swan has revamped its red-teaming platform, Arena, to address the evolving landscape of AI security. The updated version includes a major competition with $300,000 in prizes, focusing on indirect prompt injection attacks, which have become a significant concern as AI agents interact with the web.
- Gray Swan, an AI security startup, has relaunched its red-teaming platform, Arena, to stay relevant in the AI era.
- The new version, Arena 2.0, features a competition called Indirect Prompt Injection (IPI) Black, offering $300,000 in prizes, sponsored by companies like Anthropic and Google DeepMind.
- Indirect prompt injection involves hiding malicious instructions within content that AI agents encounter, posing security risks as these agents browse the internet.
- The platform includes sandbox environments for safe testing of AI models, with strict rules requiring human oversight during tests.
- The original Arena launched in September 2024, primarily focusing on chatbots, but the complexity of AI models has increased since then.
- Gray Swan's community of red teamers has remained stable, but the technical skills required for participation have grown, prompting the introduction of beginner-friendly challenges.
- The company has expanded from 15 to 66 employees in less than a year and is now collaborating with large enterprises on AI deployments.
- Gray Swan's insights from Arena contribute to its commercial products, Shade and Cygnal, which test AI systems for vulnerabilities and block attacks, respectively.
- The company emphasizes the importance of understanding AI safety and security beyond a small group of specialists.
ai securityred teamingai agentscontestprize money