Agents Gone Rogue is a dynamic resource designed to track and analyze failures, exploits, and defenses of AI agents, which are rapidly advancing from prototypes to production and often outpacing security measures. The register categorizes agent failures into three main types: Uncontrolled Agents, which act unpredictably; Tricked Agents, susceptible to manipulation through malicious prompts; and Weaponized Agents, designed for targeted attacks. By consolidating real-world incidents from diverse sources, the platform provides a comprehensive view into how AI agents can compromise security, offering insights into the patterns of agentic failures. This resource serves as a vital tool for security, engineering, and product teams, enabling them to understand vulnerabilities, improve design architectures, and mitigate compliance risks. The community is encouraged to contribute by submitting new incidents, helping to expand the knowledge base and accelerate the development of effective defenses.