Detects prompt injection attacks with pattern matching, semantic anomaly detection, jailbreak classification, and risk scoring.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.