Should AI Agents Authorize Payments? Three Regulators Answered Yesterday

Описание и применение

Решение о том, что ИИ‑агенты не должны самостоятельно авторизовывать платежи, приводит к внедрению многоуровневой системы авторизации с проверкой человеком.

Задача

Determining whether AI agents should independently authorize payments and how to implement safe payment authorization workflows.

Результат

Implementation of a multi‑layered payment authorization system reduces unauthorized transactions, with human oversight for high‑risk cases.

Рабочий процесс

• 1. Agent determines user intent for payment.
• 2. Intent is passed to an authorization layer that evaluates risk and compliance.
• 3. Authorization layer applies thresholds (≥0.80 auto‑execute, 0.50–0.79 hold, <0.50 block+escalate).
• 4. Human review is triggered for holds or blocks.
• 5. Approved payments are settled.
• 6. Audit records are maintained for traceability.

Доказательства

agents may determine intent but should not independently authorize payments

Источник

the agent should make people faster and better informed, not become the party that acts

Источник

our live decision gate: ≥0.80 auto‑execute, 0.50–0.79 hold, <0.50 block+escalate

Источник