Microsoft's security chief says frontier models, whoever builds them, cannot be trusted, and launches a product to watch them accordingly.
TechCrunch reports that Microsoft introduced Guardian Agents, a monitoring feature for the AI agents a company uses, treating them as potential insider threats rather than trusted staff. The launch, part of a program called Project Perception that Microsoft introduced in June, argues oversight should sit outside the AI labs that build the models. Microsoft's security business chief, Hayete Gallot, said frontier models, regardless of origin, cannot be trusted, not because they are malicious or easily hacked, but because they are unpredictable, while CEO Satya Nadella said it is time to stop trusting AI model providers' own safety assurances.
The launch also includes an agent-detection feature in Microsoft Defender that can identify up to 60 types of agents employees bring into the workplace, and controls that limit what data and systems an agent can reach, dynamically adjusting its permissions by task and requiring human approval if an agent appears to be making a risky decision.
Why it matters: governing the AI agents already running inside the enterprise, not just trusting a vendor's safety claims, is becoming a baseline control. Expect agent visibility, scoped permissions and human-approval gates to show up in cloud, data and security reviews next.