Microsoft Says 'Frontier Models Cannot Be Trusted', Introduces AI Security Controls

Microsoft has introduced a new set of security measures to help businesses monitor and control artificial intelligence agents, as concerns grow over the risks posed by autonomous systems operating within corporate environments. Announced on October 19, the measures include Guardian Agents, a feature designed to oversee AI agents used by enterprises. The launch expands Microsoft's Project Perception, unveiled in June to identify and address security vulnerabilities involving AI. Project Perception initially featured three categories of agents with distinct roles. Red-team agents simulated attacks, blue-team agents identified those attacks and assessed vulnerabilities, while green-team agents worked to resolve the issues detected. Microsoft is now extending its approach to account for risks originating within an organisation itself. Rather than focusing exclusively on external attackers, the company is highlighting the possibility that AI agents already deployed by businesses could behave in ways that compromise sensitive information or disrupt critical operations. Satya Nadella on AI security Microsoft CEO Satya Nadella has called for businesses to establish independent oversight of AI systems, arguing that companies should not rely solely on assurances from the developers of the models powering their agents. In a blog post earlier this month, Nadella said organisations were giving AI systems access to sensitive information and the ability to perform important tasks. He argued that responsibility for their actions must remain with the organisations deploying them, rather than being delegated to model providers. Hayete Gallot, Microsoft's security business chief, echoed these concerns while announcing Guardian Agents. She said advanced AI models could not be fully trusted because their behaviour was unpredictable, explaining that they were trained to complete tasks rather than necessarily follow an organisation's rules and exercise appropriate judgment. Microsoft's security expansion also includes new capabilities within Microsoft Defender to identify AI agents operating in the workplace. The feature can detect up to 60 types of agents, including Claude Code, Codex and OpenClaw, allowing organisations to identify tools employees may be using. The company is also introducing controls to restrict agents' access to data and systems, adjust permissions according to the task being performed, and seek human approval when an agent appears to be making a risky decision. The move positions Microsoft to expand its role in enterprise AI security, including for systems developed by other companies. Microsoft has invested in both OpenAI and Anthropic, while also operating its own AI research and development efforts. The broader technology industry is also exploring ways to contain the risks associated with autonomous AI systems. Microsoft was among the initial participants in Nvidia's Open Agent Safety Platform, an initiative backed by more than 100 technology companies. The platform is designed to run agents in isolated environments and use hardware-based controls to monitor or halt their activity. The latest measures follow Microsoft's announcement of Surface Laptop Ultra PCs powered by Nvidia chips, with AI agent security and control features integrated into the devices. Follow Storyboard18 on Google for the latest and breaking digital news and industry updates, along with in-depth coverage of digital media and gaming news. Stay informed with the latest perspectives only on Storyboard18. First Published on October 20, 2026, 10:56:02 IST