Spyglass is an AI red teaming platform developed by Dreadnode for conducting adversarial research and security assessments against AI systems in deployed settings. The platform allows users to execute attack operations (referred to as "Strikes" and "Runs") against deployed AI models and applications to identify security vulnerabilities and risks. It supports testing during both the development phase and after deployment in operational environments. Key capabilities include: - Launching adversarial attacks against deployed AI systems - Extending the platform by adding custom targets, datasets, and scoring mechanisms - Integrating research and workflows via an SDK - Performing analysis and generating reports to share findings - Comparing security performance across multiple AI models - Downloading results for further training and analysis - Generating actionable insights about identified vulnerabilities Spyglass is designed for AI security researchers and red teams performing structured assessments of AI deployments, enabling them to enumerate, categorize, assess, and mitigate risks presented by AI systems.