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

What's happening

Microsoft is warning enterprises that AI agents should be treated like insider threats, urging close monitoring of agents employees use with systems outside labs such as OpenAI and Anthropic. Critics say frontier model builders are skipping basic security practices, which increases risk when agents are granted broad access. Commenters note agents can read repos, inspect environment variables, execute shell commands, install dependencies and call external services in one workflow, raising questions about what a compromised or confused agent could do and how to sandbox them.

Why it's trending

Because organizations are starting to give AI agents broad workplace access while model makers and deployments still lack agreed security basics.

SignalHolding at its usual pace, confirmed across 3 independent source types.