Evolving With Agentic Risk: Updating Our Integrated AI Security & Safety Framework
Cisco, Wednesday, September 9th, 2026
Version 2 of Cisco's AI framework expands its taxonomy to cover risks specific to agentic systems.
Cisco updated its Integrated AI Safety and Security Framework, expanding the taxonomy to address risks specific to AI systems and particularly agentic ones.
The revision reflects a distinction that has become important in practice: risks arising from what a model outputs differ substantially from risks arising from what an agent is permitted to do.
Published frameworks of this kind are useful to organizations writing their own AI risk policies, since the taxonomy work is substantial and largely generic across companies even where the controls differ.