Microsoft says that since these agents were implemented internally, organisations have largely treated them with implicit trust, there is a need for ‘Guardian Agents’ to make matters more secure. At a time when there is a growing wave of publicly disclosed safety incidents highlighting the sheer power and potential hazards of modern artificial intelligence (AI) and AI agents, cybersecurity experts are sounding the alarm on an overlooked vulnerability: the autonomous software agents operating inside corporate networks. companies have quietly deployed thousands of internal agents to handle everyday business processes While the broader tech industry has remained fixated on external cybercriminals weaponizing models.
First announced by Microsoft in July to help cyber defenders outpace AI-driven attacks originating outside corporate walls, Project Perception is now being directed inward to oversee and govern internal agentic insiders. Built to impose deterministic controls on non-deterministic artificial intelligence, the platform uses a multi-layered cyber architecture.