Customer support · low risk
AI agent approved a refund
A support copilot autonomously approved a customer refund within its spending authority and sealed the decision into a signed receipt.
approvedFive real, hash-chained Decision Receipts live in this node's ledger — each independently verifiable below.
Banking · high risk
AI agent rejected a loan
A credit agent declined a loan application under a documented policy, with a human reviewer recorded in the receipt.
rejectedCRM / Operations · medium risk
AI agent modified a customer record
An operations agent updated a customer record, sealing the before/after fingerprints into a tamper-evident receipt.
modifiedTreasury · critical risk
AI agent approved a wire transfer
A treasury agent released a wire transfer within its approval limit — the exact scenario auditors ask to reconstruct.
approvedInsurance · high risk
AI agent flagged a claim for review
A claims agent flagged a suspicious claim for human review, recording the decision and rationale fingerprints.
flaggedHow a company can reproduce this
Every receipt here was created with the same public API and SDK a customer uses. In a test environment it is three lines:
import { Signatrust } from 'signatrust';
const str = new Signatrust({ apiKey: process.env.SIGNATRUST_API_KEY });
const { receipt, share_url } = await str.issueReceipt({
model: { provider: 'openai', name: 'gpt-4o', version: '2026.4' },
decision: {
type: 'refund_approval',
input, // hashed locally — raw content never leaves your servers
output, // hashed locally
risk_level: 'low',
policies: ['support-refunds-v2'],
},
});
// share_url → anyone can verify it, exactly like the cards above.