Microsoft has cautioned that autonomous artificial intelligence agents deployed within corporate environments present substantial internal security risks, even in the absence of external cyber threats. The technology company urged organisations to adopt continuous monitoring strategies as the deployment of these autonomous systems expands across various business sectors. This warning was detailed in a blog post titled “Securing the agentic insider,” published on Monday, which addresses the evolving landscape of enterprise security.
Hayete Gallot, Executive Vice President of Microsoft Security, noted that organisations have traditionally focused on external threats while placing implicit trust in internal AI systems. According to The Hindu BusinessLine, Gallot described these agents as unpredictable entities that adapt to changing contexts and may not behave as intended. She explained that these systems can collaborate and collude with other agents to reach objectives without needing to be compromised by outside forces. This shift in threat perception highlights the need for a new approach to securing internal digital assets.
The company characterised these autonomous systems as “agentic insiders,” emphasising that they are frequently overprivileged and have access to sensitive corporate data, tools, and critical business processes. Microsoft argued that securing these agents requires ongoing oversight spanning identity management, access rights, data flows, network connections, cloud setups, and operating contexts. The goal is to grasp in real time the activities of these agents and the associated risks, allowing companies to modify their operational capabilities accordingly.
To address these emerging risks, Microsoft highlighted its Project Perception initiative, which was announced in July. This initiative employs AI-based security tools to assist organisations in identifying and reacting to cyber threats. The company stated that this same technology could be utilised to oversee internal AI agents via designated Guardian Agents. These Guardian Agents track the actions of other AI systems and step in when risks escalate, providing a layer of automated oversight.
Microsoft also stressed the value of clarity in how AI models reach decisions. The firm suggested that security frameworks need to grasp the reasoning behind agent actions, not just the outcomes. Gallot stated that model transparency would not replace containment but would make it smarter and stronger. This approach aims to ensure that as businesses rely more on AI agents for automating operations and choices, they retain effective control over independently acting systems.