Customer-support agent Duckie writes directly to Stripe billing; discloses no insurance or liability terms for its own errors

Duckie, an AI agent platform marketed to customer-support teams, authorizes its agents to change a customer's subscription and process refunds directly in Stripe, the company's website shows, while publishing no insurance, liability cap or outcome-linked payment term covering what happens when those actions go wrong.

The company's own example shows an agent executing a billing change with no human review step described for that action: "When a customer asks to change their plan, update it in Stripe and confirm with them once it's done." A dollar threshold gates refunds instead: "Refund > $200 → human approval." Below that line, the action proceeds on the agent's own authority.

Duckie tests proposed agent behavior against historical cases before deployment: "96% match Replaying 2,190 real tickets." The company describes a policy layer applied across its whole fleet of deployed agents: "Policy engine enforced on 100% of agents."

On third-party verification, Duckie's security page says the audit it is pursuing is not yet complete: "We are actively working in the process of obtaining SOC-2 compliance. We have partnered with Thoropass, a reputable compliance automation platform, to support us." The same page addresses liability for data breaches, not for agent errors: "Duckie's contract and terms of service outline the liability in the event of a data breach or compromise of the customer's environment."

Neither page names an insurance carrier, a liability cap tied to agent errors, a customer indemnification clause, or a payment term under which Duckie is paid based on the accuracy of what its agents do. A full-text search of Duckie's homepage (4,015 characters, fetched Aug. 2 from https://www.duckie.ai/) found no occurrence of "insurance" or "indemnif." A full-text search of its security page (15,570 characters, fetched Aug. 18 from https://www.duckie.ai/security) found no occurrence of "insurance," "indemnif" or "warranty."

AI Dispatch's trust-gap rubric scores five dimensions on a 0-3 scale: runtime governance, audit trail, compliance, outcome settlement, and insurance/indemnity. Duckie's disclosed approval gate and test-replay process are more specific than many agent platforms AI Dispatch has reviewed. The gap sits in compliance, which is pending rather than certified, and in outcome settlement and insurance/indemnity, where the company currently publishes nothing.