Microsoft announces oversight of AI agents. "You can't trust"

Microsoft is implementing Guardian Agents to monitor AI agents as part of Project Perception, aimed at enhancing security. Hayete Gallot emphasized the unpredictability of advanced AI models, which are trained for results rather than adherence to organizational rules. The project seeks to establish identity and permissions for AI agents, while Guardian Agents will oversee their actions and respond to rising risks. Satya Nadella highlighted the necessity of limited trust in AI models, stressing that companies must retain responsibility for AI actions. Project Perception, announced in July, employs teams of agents for security testing, threat detection, and response, now extending to AI agent control.