Most operating systems assume the human is the one doing things. That worked fine until AI agents started acting inside your accounts, your wallets, your infrastructure.

The problem isn't that agents are insecure. It's that they inherit ambient permissions designed for humans. Every app, every session, every API key gets the same authority whether a human or an AI triggered it.

Message-passing microkernels solve this by making every action an explicit request that can be inspected, approved, or denied. No ambient access. No implicit trust. That's the design behind KeyOS.

https://foundation.xyz/blog/the-ideal-os-for-ai-is-a-message-passing-microkernel