Agentic AI: Managing Risk with Limited AI Governance

• Agentic AI is rapidly evolving and expanding, but governance is still limited. 
• Agentic AI poses a unique threat that can’t easily be addressed today.
• Microsoft is offering a degree of agentic AI governance, but customers must pay more, and race to keep up.

The rapid sprawl of agentic AI (AI agents) within organizations threatens to create a new category of risks that’s complex to resolve. Simplistically, instead of the chat-based interface associated with large language models (LLMs), agents are intended to address a human-defined need by independently determining the steps and then the technology required to solve it, and reasoning and iterating through them until the task is either resolved or determined to be unresolvable using the tools and information available to it. (For a more comprehensive definition of agentic AI and agents, see the sidebar “Agentic 101.”) 

Agentic AI offers potentially boundless interconnectivity, but this non-standardized range of AI technologies is evolving faster than most organizations can keep up with, and agentic AI continues to rapidly proliferate or come into consideration. Combine this with machine-driven agent “reasoning” using non-deterministic AI, and the ability for agents to complete a task autonomously without further human approval or intervention, and the potential risks created by agentic AI expand in unpredictable ways.