Glossary
Policy enforcement point (PEP)
A check between an agent and its tools that decides, before each call runs, whether it is allowed: who is asking, for what, with which arguments.
OWASP recommends a pre-execution policy enforcement point that validates intent and arguments and enforces schemas and rate limits, with each privileged step re-verified by a central policy engine. It also advises keeping that engine outside the agent, separating planning from execution, so a corrupted plan cannot trigger harmful actions.