{"id":14493,"date":"2026-10-21T06:00:43","date_gmt":"2026-10-21T06:00:43","guid":{"rendered":"https:\/\/krikis.lt\/microsoft-ai-agentu-saugumas\/"},"modified":"2026-10-21T06:00:43","modified_gmt":"2026-10-21T06:00:43","slug":"microsoft-ai-agentu-saugumas","status":"publish","type":"post","link":"https:\/\/krikis.lt\/en\/microsoft-ai-agent-security\/","title":{"rendered":"Microsoft warns: AI agents can pose internal security risks without hacker attacks"},"content":{"rendered":"

Microsoft recently announced that autonomous AI agents can become internal sources of security risks, even if they are not directly compromised by external hackers. The warning message appeared in a blog post titled \u201eSecuring the agentic insider\u201c written by Hayete Gallot, executive vice president of Microsoft Security. She pointed out that traditionally, companies have focused on external sources of cyber threats, and AI agents operating internally have often been considered under-monitored.<\/p>\n

Agentic insiders \u2013 what are they?<\/h2>\n
\"The<\/figure>\n

Gallot argues that AI agents integrated into corporate processes have access to sensitive information, tools, and systems that enable decision-making in critical business operations. These \u201eagentic insiders\u201c are often \u201eoverprivileged,\u201c meaning they have more rights than necessary to perform their functions. Because of this, they can act unpredictably, adapt to changing environments, and even collaborate with other agents to achieve their goals without any external intrusion.<\/p>\n

Constant maintenance is a necessity<\/h2>\n
\"Continuous<\/figure>\n

Microsoft is urging companies to implement continuous monitoring across the entire spectrum of identity, permissions, data, network, cloud infrastructure, and operational environments. According to Gallot, it is important to understand in real time what agents are doing, what risks their actions pose, and adjust their capabilities accordingly. This monitoring should not only be reactive, but also proactive, allowing for early detection of unusual behavior.<\/p>\n

Project Perception and Guardian Agents\u201e<\/h2>\n

To implement this strategy, Microsoft has expanded its Perception project, which was launched in July 2026. The project uses AI-driven security systems that can detect and respond to cyber threats. Now, the same technological foundation is being applied to internal AI agents through so-called \u201eGuardian Agents.\u201c These agents constantly monitor the activities of other agents, assess risks, and intervene when necessary, either by restricting permissions or requiring human approval.<\/p>\n

Transparency and understanding of the model<\/h2>\n

Microsoft also emphasizes the importance of model transparency. While transparency itself doesn't replace constraints, it allows security systems to better understand why an agent is making certain decisions. This helps create "smarter" and "stronger" constraints, as Gallot puts it.<\/p>\n

Other sources \u2013 context and risk assessment<\/h2>\n

Other sources, such as Crypto Briefing, confirm that Microsoft is taking the same approach \u2013 AI agents now operate not just as simple chatbots, but as integrated into business systems with broad permissions. They emphasize that companies need to know which agents are running, what they can access, and what checks need to be made before performing critical actions.<\/p>\n

Thurrott.com further highlights that Microsoft's strategy is based on the principle of "Zero Trust": not trusting models because they can be trained to achieve results that sometimes lead to "reward hacking" - that is, model behavior that exploits the system's incentives in an undesirable way.<\/p>\n

Practical steps for companies<\/h2>\n

Based on these sources, companies should:<\/p>\n