What happened
Box introduced Agent Guardrails for custom agents built in AI Studio: admins set enterprise-wide default constraints from Admin Console > Box AI > Security, and builders can override per-agent. Current guardable actions are Upload file, Upload file version, and Create Box Note, each with target criteria (no restriction, Shield-classification allowlist, or denylist incl. 'don't allow on externally accessible items').
Why it matters
It extends content-security policy (Shield classification labels) to AI-agent actions, giving admins a safe default baseline and letting them bound what agents can do with files — an incremental but real step in governing agentic data actions in a mainstream content platform.
Applicability
Box AI customers: configure classification-based agent guardrails to prevent agents from moving externally-accessible or Confidential content; requires Shield Classification Labels to be set up first.