Microsoft says AI agents are like insider threats as ‘frontier models can’t be trusted’‎

Microsoft wants enterprises to closely watch the agents employees use with systems not controlled by AI labs like OpenAI and Anthropic.