Hayete Gallot, executive vice president of Microsoft Security, said in a blog post that companies often focus only on avoiding external cyberattacks.

Microsoft AI Alert

Artificial intelligence (AI) agents operating within companies can pose a significant security risk even without being hacked. Tech giant Microsoft has warned companies that as the use of AI in offices increases, it's crucial to constantly monitor these autonomous AI systems.

The 'Agentic Insider': The New Inside Threat

Hayete Gallot, executive vice president of Microsoft Security, explained in a blog post that companies often focus solely on protecting themselves from external cyberattacks and blindly trust the AI ​​agents operating within their systems, which can prove dangerous.

According to Gallot, "The behavior of AI agents is unpredictable. They adapt as the environment changes. Even if they are not hacked, they can work together with other AI agents to achieve their goals and act arbitrarily."

Why is the risk increasing?

Nowadays, AI agents have access to confidential and sensitive corporate data, critical tools, and systems. They make decisions on their own and handle major business tasks.

Microsoft has labeled these agents "agentic insiders," meaning "insider threats." The company says these AI systems are often given excessive powers, significantly increasing security risks.

How will we get relief from this danger?

Microsoft says that to avoid this threat, the identity, rights, data and network of AI agents will have to be monitored 24 hours a day in real-time.

To address this issue, Microsoft mentioned its "Project Perception." Under this initiative, the company will use "Guardian Agents." These Guardian Agents will monitor the actions of other AI agents and stop them as soon as they detect a threat. Microsoft also emphasized that understanding why an AI model makes a decision is crucial for security.