The AI Risk Matrix: Evolving AI Safety and Security for Today

As AI models evolve from simple responders to autonomous agents, traditional safety testing is no longer sufficient. A new framework, the AI Risk Matrix 2.0, is proposed to better assess risk by adding a third dimension of "Model Agency" to the existing axes of User Intent and System Responsibility. This new dimension categorizes systems into three tiers of increasing risk: bounded Tools, adaptive Agents, and complex Collectives of multiple agents. The framework emphasizes that the greatest threats lie not just in the model but in the entire system, highlighting the need for a structured approach to manage the emergent risks of increasingly capable AI.