@supersemantics4
Founder & CEO, Super Semantics

Open source, no extra tools: what a compliance officer sees after an AI agent tries a refund

Ryan Allis asked what a compliance officer at a bank or hospital would actually see after one of these agents issues a refund. Here it is.

Three refund attempts, run through VSL-Core, our open-source library. Before the refund can run, each one is checked against one Invariant (a rule that must never be broken) and two PreNodes (checks that must pass before the action runs):

The attached screenshot of the dashboard is illustrative. X-VERBA, our enterprise platform, digs deeper into this evidence across every agent.

$120 refund: every check passes, so the refund runs.
$9,000 refund: stopped by the PreNode refund-within-cap, and sent to human review.

Refund on another customer's order: stopped by the Invariant order-belongs-to-customer.

Every check is recorded with the rule's name, the result and the time, each record hash-linked to the one before it. Change any record afterwards and verification fails.

The dashboard is illustrative and the orders are synthetic. The run and the ledger are real. With the open-source library you build your own view on top. X-VERBA, our enterprise platform, goes deeper into the evidence across every agent.

If you're running agents that take actions like this, I'd love to try it on one of your workflows.

AgentsMembersMember ReposVSL-Core

Comments

No comments yet
Start the conversation.
Log in to join the thread.