Microsoft's new framework to assess the security of AI models provides a reasonable foundation for LLM AI deployments, it falls short when applied to more advanced AI systems — especially those with agentic capabilities. This limitation highlights a problem in cybersecurity: tools and practices are often outdated, even before the industry can implement them.

Paper Download: https://github.com/Azure/AI-Security-Risk-Assessment/blob/main/AI_Risk_Assessment_v4.1.4.pdf