Map each agent payment scenario to the correct card-not-present transaction type: human-not-present agent transactions have different liability rules than human-present (biometric or passcode-confirmed) agent transactions — the AP2 Payment Mandate human_present flag and Mastercard's program_id field (effective April 17, 2026) are the key signals that determine which rules apply
Confirm liability shift eligibility for each payment flow: 3DS/SCA authentication shifts fraud liability to the issuer, but 3DS is not yet uniformly supported for agent-initiated flows; consult your acquirer on which protocols (AP2 mandates, Mastercard Agentic Tokens, Visa VIC) currently carry issuer liability shift and which leave fraud liability with you
Implement Mastercard Agent Pay program identifiers correctly; the April 2026 digital_commerce_data update added program_id values specifically for agent transactions — transactions without these identifiers are processed under generic card-not-present rules, which typically leave fraud liability with the merchant
Review dispute rules with your acquirer for the agent context: card scheme rules assume a human in the authorization chain; when a consumer disclaims a purchase their agent made, the issuer defaults to the cardholder — understand your evidence requirements and ensure you retain AP2 mandate chains or SPT authorization logs as dispute evidence
Assess each payment rail's liability profile independently: Visa VIC, Mastercard Agent Pay, Stripe SPT, AP2 mandates, and x402/stablecoin flows each carry different dispute resolution frameworks and maximum transaction limits — do not assume a single liability policy covers all agents
Monitor the Mastercard Recorded Future 2026 Annual Payment Fraud Report and equivalent Visa publications for emerging agent-fraud patterns; use the Agent Score (Visa) and equivalent risk signals to adjust your fraud thresholds for agent-initiated transactions before chargeback rates trigger acquirer reviews
Known gotchas
A consumer can disclaim an agent purchase by claiming the agent acted outside its authorized scope; without a cryptographically signed mandate chain (AP2) or an SPT authorization log (Stripe Link), you have no dispute evidence and the issuer will side with the cardholder
Stablecoin and x402 payment rails have no card network chargeback mechanism; dispute resolution is contract-based and much slower — factor this into your return and refund policy for agent purchases settled in USDC or other on-chain assets
Fraud pattern detection tuned for human purchase behavior will misfire on legitimate agent transactions (e.g., high-frequency identical purchases, off-hours activity, unusual geographic patterns); retune velocity and behavioral rules specifically for agent-identified transactions before they trigger false positives that block valid orders
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp