Last released Sep 25, 2026
Runtime security for AI agents — scope enforcement, prompt-injection screening, output redaction, human approvals, cost circuit breakers, honeypots.