How to move from AI discovery to AI enforcement (www.techradar.com)

Many enterprises have completed the initial discovery phase of their AI programs, identifying a surprising number of AI agents in use. However, these programs often stall at this stage, as organizations struggle with implementing effective enforcement. This is significant for the AI/ML community because understanding not just what AI tools are in place, but also how to manage and control their usage, is crucial for security and compliance. With the average enterprise running 37 AI agents, more than half of which lack security oversight, the challenge becomes clear. Enforcement refers to the mechanisms that can intervene in real-time during AI operations, such as blocking actions, scoping access, gating actions behind approval, or terminating processes. Each method comes with its complexity and implications. Traditional methods like network blocking have proven inadequate, as they often fail to address local models and embedded tools that don’t communicate over standard network channels. The emerging need for refined enforcement strategies highlights the importance of developing separate controls for unauthorized agent actions and privilege enforcement. Experts emphasize that effective enforcement must operate at the moment actions are executed, ensuring organizations can respond decisively and accurately to potential risks associated with AI systems.