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.