For calls routed through the AI gateway, Yardstick can check what comes back. Set rules for PII (emails, SSNs, card numbers, key shapes), a custom regex, or a JSON shape, each in warn or block mode. Warn logs the violation & passes the response through; block replaces it with an error so the bad output never reaches the caller. Every violation is logged (the rule & a match count, never the matched content), so guardrails are enforced & auditable, not just observed.
Output guardrails
Block or flag bad model output at the gateway: PII, secrets, regex & JSON shape.
