Microsoft on Monday unveiled new security features aimed at protecting businesses from the growing risks posed by AI agents, as the company continues to warn organisations about potential security threats associated with the technology.
One of the key additions is Guardian Agents, a feature designed to monitor the AI agents deployed within an organisation.
The tool is part of Project Perception, which Microsoft introduced in June.
Project Perception was Microsoft’s response to advanced hacking models, including Anthropic’s Mythos and Fable, as well as OpenAI’s Daybreak system for scanning security vulnerabilities.
When it launched, Project Perception featured three categories of AI agents. The “red team” simulated hackers by launching attacks, while the “blue team” detected these attacks and assessed the vulnerabilities identified.
The “green team”, meanwhile, was responsible for fixing the security weaknesses.
The product was initially designed on the assumption that AI-powered attacks would originate from outside an organisation.
However, Microsoft CEO Satya Nadella has recently warned that the threat could also come from AI agents deployed and paid for by businesses themselves.
The concern mirrors the risk posed by rogue employees, commonly known as an “insider threat”, who exploit their access to an organisation’s systems and data.
In a private blog post, Nadella warned enterprises against placing complete trust in AI model providers, even when they offer assurances about the safety and security of their agentic products.
He urged businesses to deploy independent systems, separate from those controlled by AI developers, to monitor and control the agents launched by their products.
“We are deploying these complex agentic systems and models, with access to our most sensitive data and giving them the ability to take mission-critical actions on our behalf!” he wrote. “We simply can’t outsource responsibility for what intelligence does on our behalf. A model provider’s assurances do not relieve us of that responsibility. … In other words, we need to separate the supply of intelligence from the authority over it.”
Microsoft’s head of security business, Hayete Gallot, was even more direct in her assessment.
“Unfortunately, frontier models, regardless of their origin, cannot be trusted,” she wrote, adding that this isn’t because they are malicious or easily hacked, but because they are unpredictable. They have been trained to achieve tasks, not follow “organizational norms or judgment.”
