|
Artificial intelligence (AI) agents deployed within companies could pose significant internal security risks even without being compromised by hackers, Microsoft has warned, calling for continuous monitoring of autonomous AI systems as their use across enterprises grows.
In a blog post titled "Securing the agentic insider" published on Monday, Hayete Gallot, Executive Vice President, Microsoft Security, said organisations have traditionally focused on external cyber threats while placing implicit trust in AI agents deployed within their own systems. "Agents are unpredictable. They will adapt as the context changes, may not behave as intended and will collaborate and collude with other agents to reach their objectives, without needing to be compromised," Gallot stated. Microsoft warned that AI agents increasingly have access to sensitive corporate data, tools and systems, allowing them to make decisions and take actions in critical business processes. The company described these autonomous systems as "agentic insiders", highlighting the security risks arising from their growing role within organisations. "These agents are frequently overprivileged, have access to data, tools and systems and can take actions and make decisions in critical processes. They have become active participants in the enterprise," Gallot said. The warning comes as companies increasingly use AI agents to automate business operations and decision-making, raising concerns about how organisations can maintain oversight of systems capable of acting independently. Microsoft argued that securing these agents requires continuous supervision across identity, permissions, data, networks, cloud infrastructure and operating environments. "Monitoring and securing those agents must become continuous. We need to understand in real time what the agents are doing and what risks those actions create, and we need to adjust their ability to act accordingly," Gallot stated. To address these risks, Microsoft highlighted its Project Perception initiative, announced in July, which uses AI-driven security systems to help organisations detect and respond to cyber threats. The company said the same technology could be used to supervise internal AI agents through so-called Guardian Agents, which monitor their actions and intervene when risks increase. Microsoft also emphasised the need for greater transparency into how AI models make decisions, arguing that security systems must understand not only what agents are doing but also why they are taking particular actions. "Model transparency will not replace containment. It will make containment smarter and stronger," Gallot said. (ANI)
|
