Microsoft has unveiled new security tools designed to protect businesses from the unpredictable behavior of AI agents, infohub.kz reports.
Under its Project Perception platform, the company has released a Guardian Agents feature to monitor what artificial intelligence does. Microsoft CEO Satya Nadella said companies can no longer blindly trust the developers of AI models. Autonomous agents gain access to confidential data and can carry out critical tasks, which turns them into the equivalent of a rogue employee — an insider threat.
Hayete Gallot, who heads Microsoft's security division, called "frontier models" unpredictable: they are trained to complete tasks, not to follow corporate norms. To address the problem, Microsoft proposes separating the supply of intelligence from control over it. The new Defender features can identify up to 60 types of third-party agents, restrict their access to systems and require human confirmation when risks arise.
The corporation has also joined Nvidia's Open Agent Safety Platform, which allows agents to be isolated and their operations to be blocked at the hardware level in an emergency.