Production blocking is a security control that stops a risky action inside a live system instead of only logging it or sending an alert. In practice, the control sits at execution time and can deny a tool call, cancel a transaction, reduce permissions, or require extra checks before the operation reaches production data or infrastructure.
This matters because many attacks succeed only when a dangerous step is allowed to run. In agentic AI, for example, a compromised prompt or malicious instruction can push an agent to access files, call APIs, or modify workflows. Production blocking helps contain that behavior by enforcing policy in real time. Defenders use it to limit blast radius, support least privilege, and prevent harmful changes before they become incidents. Its value depends on accurate policy decisions, low latency, and careful tuning to avoid blocking legitimate work.