When Software Spends: Payments Build an Authorization Layer for AI Agents
Networks and banks are defining how autonomous agents get payment credentials, spending limits, and someone to blame.
Telenor via Wikimedia Commons · CC BY-SA 3.0AI agents that book, order, and procure need a way to pay, and the payments industry has concluded that handing them a stored card number is not it. Networks, banks, and standards groups are building an authorization layer for machine spending: scoped credentials, per-agent limits, and cryptographic proof of which agent initiated what under whose authority.
The design center is accountability. Merchants want assurance that an agent transaction will not become a dispute; cardholders want spending they can bound and revoke; issuers want liability rules written before the volume arrives rather than litigated after.
Early deployments live in controlled corridors, corporate procurement, travel rebooking, subscription management, where the counterparties are known and the limits are tight. The participants describe the sequencing as deliberate: teach software to spend small before trusting it to spend at all.