Detects jailbreak and prompt injection attempt patterns in input text.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.