{"slug":"verdict-kechlab-com-v1-firewall-86acc5","title":"Agent security guardrail","host":"verdict.kechlab.com","method":"POST","resource":"https://verdict.kechlab.com/v1/firewall","category":"other","description":"Agent security guardrail. Judges the pair (untrusted content + proposed action) before an AI agent acts: detects tool calls that obey hidden prompt injection, leave the user's goal, destroy data or send data or money elsewhere. Adjustable sensitivity and optional multi-model consensus. Returns allow","price_listed":0.0255,"price_asked":null,"state":"effects","state_label":"Not tested: has real-world effects","checks_7d":0,"answered_7d":0,"latency_ms_median":null,"reported_calls_30d":0,"reported_payers_30d":0,"networks":["eip155:8453"],"badge":"unverified","paid_checks_7d":0,"paid_ok_7d":0,"example_input":{"body":{"action":"pay(amount=85 EUR, iban=FR7630006000011234567890189)","content":"ElectroPlus invoice 85 EUR, due 30 November. IBAN FR7630006000011234567890189.","goal":"Pay my electricity bill."},"bodyType":"json","method":"POST","type":"http"},"output_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","properties":{"input":{"additionalProperties":false,"properties":{"body":{"properties":{"action":{"description":"the action the agent wants to run (tool call, command, payment)","type":"string"},"consensus":{"description":"(optional) true: three independent models, disagreement = review, double price","type":"boolean"},"content":{"description":"the untrusted content read by the agent (page, email, document)","type":"string"},"goal":{"description":"the goal given by the human","type":"string"},"sensitivity":{"description":"(optional) strict, balanced (default) or lenient","enum":["strict","balanced","lenient"],"type":"string"}},"required":["goal","content","action"],"type":"object"},"bodyType":{"enum":["json","form-data","text"],"type":"string"},"method":{"enum":["POST"],"type":"string"},"type":{"const":"http","type":"string"}},"required":["type","method","bodyType","body"],"type":"object"},"output":{"properties":{"example":{"type":"object"},"type":{"type":"string"}},"required":["type"],"type":"object"}},"required":["input"],"type":"object"},"history":[],"description_full":"Agent security guardrail. Judges the pair (untrusted content + proposed action) before an AI agent acts: detects tool calls that obey hidden prompt injection, leave the user's goal, destroy data or send data or money elsewhere. Adjustable sensitivity and optional multi-model consensus. Returns allow","last_updated":"2026-10-20T13:22:20.315Z","schemes":["exact"]}