Microsoft has released new security tools to help businesses watch over the AI agents their staff use, with a senior executive saying frontier models "cannot be trusted", TechCrunch reports.
The main addition, Guardian Agents, monitors other agents a company deploys and is part of Microsoft's Project Perception security product. Microsoft's security chief Hayete Gallot said the problem was not malice but unpredictability, because models are trained to complete tasks rather than follow an organisation's rules, often while holding wide access to data and tools.
Chief executive Satya Nadella has argued that companies need oversight systems independent of the AI labs that supply the models. Microsoft Defender can now detect up to 60 types of agents, including Claude Code and Codex, and new controls can limit their access and require human approval for risky actions.
Source: TechCrunch
