Blog

Agent-initiated payments are real now. What it means for your fraud stack

The card networks and payment platforms are shipping rails for AI agents to pay on a user's behalf. The new liability and trust questions, and the signals that still hold up.

Through 2026 the payment industry stopped treating AI agents as a novelty and started building rails for them. The major card networks and payment platforms now offer ways for an agent to initiate a payment on a person's behalf, with protocols for passing delegated authority along with the transaction. This is a genuine infrastructure shift, and it lands a set of new questions squarely on fraud teams who did not ask for them. If you take payments, this is heading your way whether or not you are ready.

The questions the new rails open

  • Did the person really authorise this? An agent payment is a payment-by-proxy. The hard part is confirming the agent is acting on genuine, delegated authority rather than running loose or hijacked.
  • Who is liable when an agent gets it wrong? An agent that buys the wrong thing, or is manipulated into a purchase, creates a dispute that does not fit neatly into existing chargeback categories.
  • How do you tell a trusted agent from a hostile one? The same payment rail that a legitimate agent uses is exactly what a malicious one will try to use too.

Why behavioural and device signals weaken here

Much of the modern fraud stack leans on the person being present: device fingerprints, behavioural biometrics, the rhythm of a real human checking out. Agent payments remove the human from that moment by design. The signals that assumed a person at a keyboard degrade precisely when you need them. The industry's answer — and I agree with it — is to lean on provenance and reputation: not "how did this session behave?" but "where did it come from, and does that source have standing?"

The signals that still hold up

Two checks survive the shift because they do not depend on a human being present, and they are cheap enough to run on every transaction:

  • The network and its reputation. A trustworthy agent operates from identifiable, stable infrastructure. A payment initiated through a residential proxy, a VPN or anonymising hosting is carrying a flag no delegated-authority token can paper over.
  • The issuing country behind the card. Read from the BIN, it is a fast corroboration that the payment's origin lines up with the rest of the order.

Get ahead of it

You do not need to solve agent liability to protect yourself today. You need to know, for every payment, what network it came from and whether that network is one a legitimate agent would use. Fraudex returns that verdict — network, type, bot flag — plus BIN intelligence in a single call, so agent-initiated and human payments run through the same reliable check. See the proxy detection API, the VPN detection API and the BIN lookup API.