The verdict before your AI agents act
One gate for your AI agents.
Your policies. Full audit.
Approve or escalate before agents touch refunds, email, CRM, or data—then full audit for compliance and insurers.
Try it now — live
Built for CISOs and compliance. Insurer-ready.
The problem
When your AI agents run refunds or touch sensitive data, there’s often no single gate—actions run without a check, or logic is scattered. When something goes wrong: who approved it? Where’s the log? Auditors want one trail: who did what, when, and whether a human said yes. Verdict is that gate—and that trail.
How it works
Your AI agent calls the Verdict API
Before a risky action, your agent (or bot, script) sends actor, action type, and context to our authorize endpoint.
We evaluate
Deterministic rules—no LLM. Same input, same result. In policy → approved. Outside → human queue.
Every decision logged
Hash-chained audit. Export CSV, PDF, or VC for regulators and insurers.
curl -X POST "verdict-production-a904.up.railway.app/authorize_action" \
-H "Content-Type: application/json" \
-d '{"actor_id": "agent_123", "action_type": "refund", "metadata": {"amount": 200}}'
# → APPROVED + proof_of_authority or ESCALATED + request_idSDK quickstart: from verdict import VerdictClient; client = VerdictClient(); client.authorize("agent_123", "refund", {"amount": 200}) (Python). Node: import { VerdictClient } from "verdict-sdk".
See it in action
Three steps: risky actions intercepted, human approval in one place, provable audit trail.
Why it matters
Real incidents. One gate could have stopped them.
Air Canada paid out after a chatbot promised a refund it couldn’t honor. Automations have sent the wrong data. One approval step—before your agent acts—would have changed the outcome. Verdict is that step.
Who it’s for
Teams that can’t afford a single bad agent decision.
CISOs, compliance, and product leads whose AI agents touch refunds, email, CRM, tickets, or sensitive data.
Fintech
Refunds & payouts
Healthtech
Comms & records
SaaS / Support
Tickets & CRM
Operations
Bots & scripts
You’ve seen how it works. Try it.
Run the demo (simulation, escalation queue, audit log) or get started with a pilot.