The 3Cs: A framework for AI agent security

No developer would leave their laptop unattended and unlocked. The risk is obvious. Root-level access to production systems, repos, databases, credentials, APIs. Anyone who sat down could review PRs, modify files, commit code, and access anything the developer can access.

Yet this is how many teams are deploying agents today.

Every time execution models change, security frameworks need to change with them. Agents force the next shift. They execute at machine speed, in parallel, at a pace no human can follow. More approvals and more alerts don't solve the problem. When agents run hundreds of actions in parallel, warnings become noise.

The fix is governance built into the execution layer. Docker's security team put together a framework for thinking about this clearly: the 3Cs. Contain the blast radius, starting with Docker Sandboxes, which runs every agent session inside a microVM-isolated environment with a hard boundary at the process level. Curate what the agent can see. Control what happens at runtime, before actions complete. This framework maps the governance problem to the three things any serious solution has to do.