Which best captures the key idea of "Guardrails for agents"?

ASUM, AVERAGE, COUNT, MIN, MAX handle most everyday calculations BAlways sanity-check AI's math and claims against the raw data CAsk the model to check its answer or find its own mistakes before finalising DLimit tools, add approvals, and cap steps to keep agents safe