Microsoft has unveiled new security tools designed to protect the corporate sector from threats posed by artificial intelligence agents. The key announcement was the launch of Guardian Agents, a special monitoring system that controls the behavior of autonomous AI assistants within companies.
Microsoft has unveiled new security tools designed to protect the corporate sector from threats posed by artificial intelligence agents. The key announcement was the launch of Guardian Agents, a special monitoring system that controls the behavior of autonomous AI assistants within companies.
According to TechCrunch, the new tools are part of Project Perception, a comprehensive solution launched by Microsoft in June 2026 to counter threats in the field of artificial intelligence. The launch came amid growing concerns that autonomous agents could pose risks similar to the actions of unreliable or unscrupulous employees.
Microsoft CEO Satya Nadella said it was time for businesses to stop blindly trusting the assurances of AI model vendors, and stressed the need to implement independent oversight systems that are not directly accountable to development labs.
“We deploy these complex agent systems and models, giving them access to our most sensitive data and the ability to perform critical actions on our behalf. We simply cannot outsource the responsibility for what AI does on our behalf. The assurances of the model provider do not absolve us of this responsibility. In other words, we need to separate the provision of intelligence capabilities from the authority over them,” Nadella noted.
Microsoft’s head of security, Gaëta Gallotte, also highlighted the risks of using modern AI tools without proper oversight.
“Unfortunately, advanced models, regardless of their origin, cannot be trusted. This is not because they are malicious or easily hackable, but because they are unpredictable. They have been trained to perform tasks rather than follow organizational norms or common sense,” Gallot explained.
In addition to Guardian Agents, Microsoft has updated its Microsoft Defender security suite. Now the system can detect up to 60 types of third-party agents that employees can use in the workplace (including Claude Code, Codex, and OpenClaw). The update allows you to dynamically restrict AI access to corporate data and require mandatory human confirmation if an agent tries to perform a potentially risky operation. The corporation also became a participant in Nvidia’s Open Agent Safety Platform initiative, which provides isolation and hardware emergency shutdown of uncontrolled AI systems.
