Rules and checks that prevent an AI agent from doing things it should not do. Built into prompts, code, or review processes. Prompt-level guardrails are the weakest. Code-level guardrails (the agent literally cannot call the function) are the strongest.
Related terms
Authority Boundary
A clear line that defines what an AI agent is allowed to do and what it must not do. Includes tool access, decision rights, dol...
Failure Mode
A required field on every knowledge claim documenting what happens when the rule is violated. The opposite of a happy-path doc ...
Override Authority
The power to overrule an AI agent's decision or action. Defined in advance with clear conditions and escalation paths. Override...
Build with this on OTP
OTP encodes coordination intelligence so AI agent teams can run on it. If this term shows up in your team's playbook, it belongs in your OOS.
Found an issue with this definition? Tell us and we'll fix it.