BACK TO INDEX
CASE-2026-012 // VOUCH
RESTRICTEDAI Safety / Financial Transactions

Vouch

STATUS:ACTIVE
RISK:CRITICAL

EXECUTIVE SUMMARY

Vouch is a merchant‑side admission and evidence layer for AI buyers that provides deterministic ADMIT/ESCALATE/REFUSE decisions for AI‑mediated financial transactions. It generates signed hash‑chained receipts for every decision and operates as a pure engine that never contacts external gateways or relies on LLMs for decision making, ensuring full auditability and explainability of financial flows.

PROBLEM STATEMENT

AI‑mediated financial transactions currently lack trustworthy verification mechanisms. Existing systems often rely on opaque LLM‑based decisions or third‑party gateways that introduce non‑determinism, hallucination risks, and single points of failure, making it impossible for merchants to cryptographically prove transaction validity or refusal.

INVESTIGATIVE FINDINGS

  • ▸Financial regulatory frameworks require deterministic outcomes for auditability and compliance reporting
  • ▸LLM‑based decision systems introduce unacceptable non‑determinism and hallucination risks in financial contexts
  • ▸Third‑party payment gateways create centralization risks and single points of failure for transaction processing
  • ▸Merchants require cryptographic proof that they performed due diligence on AI‑initiated transactions

BEHAVIORAL ANALYSIS

Designed as a trust‑minimized system where explainability and auditability are prioritized over convenience. Every decision is fully cryptographically verifiable without relying on black‑box AI components, ensuring merchants can prove compliance with financial regulations.

TECHNICAL EXECUTION

Built with Node.js and TypeScript using a microservices architecture. Features a rule‑based decision engine that evaluates transaction requests against configurable policy rules written in a domain‑specific language. Implements cryptographic signing using ECDSA for generating non‑repudiable hash‑chained receipts. Utilizes PostgreSQL for persistent storage of decisions and audit trails, with Razorpay test‑mode integration for simulating payment flows without real money movement.

SYSTEM ARCHITECTURE

Transaction Request Intake → Policy Evaluation Engine (Rule‑Based DSL) → Cryptographic Signing Module (ECDSA) → Audit Trail Storage (PostgreSQL) → Receipt Generation & Delivery → Optional Payment Gateway Integration (Razorpay Test Mode)

OPERATIONAL FEATURES

  • ✓Deterministic ADMIT/ESCALATE/REFUSE decisions based on configurable policy rules
  • ✓Signed hash‑chained receipts providing non‑repudiation audit trail
  • ✓Pure decision engine with zero external API calls during operation
  • ✓Support for custom policy rules via domain‑specific language
  • ✓Integration test suite with Razorpay test‑mode for safe development
  • ✓RESTful API for integration with merchant systems
  • ✓Comprehensive logging and monitoring capabilities

TECHNOLOGICAL ASSETS

Node.jsTypeScriptPostgreSQLRazorpay (Test Mode)ECDSA

CURRENT STATUS

Active development with deployed production instance. Recent work includes expanding the policy language syntax, adding webhook support for real‑time notifications, and implementing role‑based access control for administrative functions. Preparing for mainnet deployment with additional payment providers.

CONCLUSION

Provides a trustworthy layer for AI‑mediated financial transactions that delivers deterministic outcomes with full cryptographic auditability, enabling merchants to confidently process AI‑initiated payments while maintaining regulatory compliance.

EVIDENCE LINKS

REPOSITORY LIVE DEPLOYMENT

CASE METADATA

OPENED:2026-09-11
ID:CF-2026-012

TAGS

AI SafetyFinancial TransactionsDeterministic SystemsAuditabilityTypeScriptPostgreSQLCryptographyReceipt Generation
SESSION ACTIVELAST SYNC: --:-- --/TRACE:HOME/CASES/CF-2026-012//ACTIVE[ESC] to exit contextreturning may invalidate findings
○ STANDARDCLEARANCE: LEVEL 2