Microsoft launched new security features on Monday, aimed at protecting enterprises from the growing threat of AI proxies. With this release, Microsoft continues to remind enterprises to be aware of the risks posed by AI and proxies.
The most notable new feature is Guardian Agents, which is a new function for monitoring the AI proxies used by enterprises. Guardian Agents is part of Project Perception, while Project Perception is a product launched by Microsoft in June.
Project Perception is Microsoft's response to powerful hacker models such as Anthropic, Mythos, and Fable, as well as to the security vulnerability scanning system Daybreak of OpenAI. Project Perception initially introduced three types of proxies: the "Red Team" that acts as hackers to launch attacks, the "Blue Team" that detects the Red Team and evaluates the vulnerabilities they find, and the "Green Team" that is responsible for fixing those vulnerabilities.
The design of this product is based on the assumption that AI hackers will come from outside the company. However, recently, Microsoft CEO Satya Nadella has been warning that AI the enemies may actually be those proxies that enterprises have already deployed and for which they have paid, just like out-of-control human employees, what is known as "internal threats."
Earlier this month, Nadella warned on his personal blog that it was time to stop trusting AI model providers, even though they claim to provide security and protection guarantees for their proxy products. He urged companies to establish independent systems that are not controlled by AI laboratories, in order to monitor and control the proxies initiated by these AI laboratory products.
He wrote, "We are deploying these complex proxy systems and models to give them access to our most sensitive data and to empower them to take critical actions on our behalf! We cannot outsource the responsibility for representing the intelligent behavior we perform. In other words, we need to separate the provision of intelligence from its authority."
When announcing Guardian Agents, Microsoft's head of security business, Hayete Gallot, was more direct in his statement.
"Unfortunately, frontier models, regardless of their origin, cannot be trusted," she wrote. She added that this is not because they are malicious or easily hackable, but because they are unpredictable. Their training goal is to complete tasks, rather than to follow 'organizational norms or judgment.'"
In addition, these models are often granted permissions akin to those of a “kingdom key,” allowing them access to data and tools that could cause serious damage.
Microsoft's solution is both surprising and self-serving. What's surprising is that Microsoft's relationship with OpenAI and Anthropic is well-known to be very close, and they have invested in both companies. OpenAI is also one of Microsoft's largest cloud customers, and Microsoft itself has a AI laboratory for developing models. Therefore, Microsoft is essentially sowing distrust in its own business interests.
The self-interest aspect is that Microsoft hopes that companies will purchase proxy security products from it, as the cloud service provider, rather than from Anthropic and OpenAI. However, setting aside the conflict of interests, its logic is reasonable. Software that monitors proxy behavior should be independent of the proxy providers, just as auditors should be independent of the organizations being audited.
As Nadella said, "We need to separate the provision of intelligence from its permissions."
Microsoft is now seizing this business opportunity in various ways. In addition to Guardian Agents, it has also added proxy detection functionality to its Microsoft Defender product, which can identify up to 60 types of proxies that employees bring into the workplace, including Claude Code, Codex, and OpenClaw. Microsoft has also enhanced features to restrict proxy access to data and systems, and will dynamically adjust proxy permissions based on tasks, including requiring manual approval when a proxy appears to be making high-risk decisions.
The idea of treating proxies as untrustworthy insiders is also becoming popular in a wider range of industries.
Microsoft is one of the founding members of NVIDIA's newly launched Open Agent Safety Platform, a platform that has received support from over 100 other technology companies. This platform will isolate proxies in a secure environment and can monitor their activities and prevent them from functioning, essentially acting like a "shutdown switch." This follows Microsoft's previous announcement of a new Surface Laptop Ultra PC designed for proxies, which are equipped with NVIDIA chips and come built-in with proxy security and control features.
