Security boundaries in agentic architectures

What this post added

This post introduces a security architecture for agentic systems by defining four actors (agent, agent secrets, generated code execution, filesystem) and advocating for distinct security boundaries between them. It proposes separating agent compute from generated code execution using microVMs and credential brokering proxies to mitigate risks from prompt injection and credential theft. Architectures discussed include zero boundaries, secret injection without sandboxing, and separating agent compute from sandbox compute.