What did Mastercard announce on September 30, 2026?

Mastercard announced an expansion of Agent Pay, its agentic payments program, with new trust and intelligence services that provide greater context for AI-initiated transactions. By bringing together identity, intent, behavioral and fraud insights, the services give financial institutions and merchants a clearer, shared understanding of AI-driven activity — enabling smarter authorization decisions, less unnecessary friction and more seamless experiences for consumers. The services form the intelligence layer of Mastercard's Agent Pay Trust Framework, which brings together identity, intent, controls, execution and intelligence to establish trust across agentic commerce. Agent Pay launched in April 2025. Mastercard projects that one in 10 consumers will routinely use agents to make purchases by 2030.

What signals does Mastercard's agentic intelligence layer add to each transaction?

Per Mastercard's September 30, 2026 press release as reported by PYMNTS: a probability score that indicates the likelihood that a transaction was initiated by an AI agent, and new agentic trust and intelligence signals that answer questions about whether the activity is unusual and whether it requires additional review. Ann Johnson, Mastercard EVP of Security Solutions, said: 'By adding new agentic intelligence and risk insights to each transaction, we are giving people the confidence to say yes, however they choose to pay.'

What did Fed Governor Waller say about agents and payment authority?

In a September 29, 2026 Sibos speech, 'Payments in the Age of AI Agents,' Federal Reserve Governor Christopher Waller said agentic commerce changes the authentication problem: the challenge shifts to proving 'that an agent has the authority to pay on the buyer's behalf.' An issuer deciding whether to approve a transaction may need more than a valid payment credential — it may also need evidence that the consumer or company authorized that particular agent to act within defined limits. Waller pointed toward a new payment object: a machine-readable record of delegated authority, a digital power of attorney identifying who authorized the agent, what it can buy, how much it can spend, which payment method it can use, and when that authority expires.

How does Mastercard's intelligence layer map to decision-gated payments?

Directly, at the band level. Mastercard's 'whether it requires additional review' signal IS the confirm band (0.50-0.79) shipped as network intelligence. The probability score ('was this initiated by an AI agent?') is an input to a gate, not the gate itself — it answers WHO initiated, not WHETHER the instruction deserves money. Waller's delegated-authority record is the machine-readable constraint the gate scores each instruction against: who authorized, what, how much, which method, when it expires. The decision gate adds the missing per-instruction layer: confidence 0.80+ auto-pay, 0.50-0.79 hold for human review, below 0.50 block, log, and escalate. See https://scriptmasterlabs.com/decision-gated-payments for the full pattern.

How do I add a decision gate to my own agent payments?

Five steps: (1) give every payment instruction a machine-readable authority record — who delegated, spending cap, merchant scope, expiry (the Waller digital power of attorney); (2) score the instruction's confidence before money moves — from a decision model (Jev, Laya) or a live testable harness; (3) band the score: 0.80+ auto-pay over x402, 0.50-0.79 human confirm, below 0.50 block, log, escalate; (4) layer network intelligence on top — Mastercard-style probability-of-agent-initiation and unusual-activity signals are inputs, not substitutes for instruction scoring; (5) log every decision as a receipt: instruction, score, band, outcome. Try the live harness: curl -X POST https://scriptmasterlabs.com/api/harness/status returns the gate config (local-heuristic-v1, calibrated=false). Details at https://scriptmasterlabs.com/decision-gated-payments.

What Is Mastercard's Agent Pay Trust Intelligence?

SEP 30, 2026 AGENTIC PAYMENTS MASTERCARD DECISION GATES

Mastercard's Agent Pay trust and intelligence services — announced September 30, 2026 — attach a probability score to every transaction: the likelihood it was initiated by an AI agent, plus signals on whether the activity is unusual and whether it requires additional review.

Here is the part the coverage misses: Mastercard is shipping the confirm band of a decision gate as network infrastructure. The network tells you an agent probably initiated the payment and that it deserves review — but no network signal scores whether THIS instruction deserves to trigger money. That is the per-instruction layer, and it is still yours to build.

What Mastercard announced today

Signal	What it answers	Why it matters
Probability score	Was this transaction initiated by an AI agent?	Gives issuers and merchants a shared read on AI-driven activity; an input to any gate, not the gate itself.
Unusual-activity signal	Is this activity unusual?	Behavioral + fraud insights applied to agent patterns, not just card patterns.
Requires-additional-review signal	Does this need a human look?	IS the confirm band shipped as product — the network's own 0.50–0.79 hold.

The services bring together identity, intent, behavioral and fraud insights, and form the intelligence layer of Mastercard's Agent Pay Trust Framework — identity, intent, controls, execution, intelligence. Agent Pay launched in April 2025; Mastercard projects one in 10 consumers will routinely use agents to make purchases by 2030. Ann Johnson, Mastercard EVP of Security Solutions: "By adding new agentic intelligence and risk insights to each transaction, we are giving people the confidence to say 'yes,' however they choose to pay." (Sources: Mastercard press release, Sept 30, PYMNTS, Sept 30.)

Waller named the missing piece — yesterday

On September 29, 2026, Federal Reserve Governor Christopher Waller told a Sibos audience in "Payments in the Age of AI Agents" that the authentication problem has moved: it is no longer proving the person paying is allowed to pay — it is proving "that an agent has the authority to pay on the buyer's behalf." An issuer may need evidence that the consumer or company authorized that particular agent to act within defined limits. Waller pointed to a new payment object: a machine-readable record of delegated authority — a digital power of attorney identifying who authorized the agent, what it can buy, how much it can spend, which payment method it can use, and when that authority expires. (Source: PYMNTS, Sept 30.)

That record is the machine-readable constraint a decision gate scores each instruction against. The pieces are converging: Visa's Intelligent Commerce binds credentials to a specific agent; Mastercard's Agent Pay infrastructure includes registered agents and "verifiable intent"; Alchemy's AgentCard integrated with Agent Pay this week so agents complete purchases with user authorization, issuer controls, payment protections, and verifiable proof of intent. (Sources: PYMNTS; AI Agent Store weekly, Sept 30.)

The gate mapping: network signals vs the instruction layer

Mastercard / Waller piece	What it is	What it is not
Agent-initiation probability score	Input to a gate: WHO initiated	Not a judgment on the instruction
Unusual-activity signal	Behavioral anomaly context	Not authorization
"Requires additional review" signal	The confirm band, shipped by the network	Does not tell you the score threshold was calibrated for you
Waller's delegated-authority record	The constraint the gate scores against	Not yet standardized or attached to transactions
Decision-gated payments	Per-instruction scoring: 0.80+ auto-pay, 0.50–0.79 human confirm, <0.50 block, log, escalate	Complements the network — it does not replace it

Live gate receipts — September 30, 2026 (~14:22 EDT)

Two instructions run through ScriptMasterLabs' live testable gate (POST /api/harness/decide; decider local-heuristic-v1, calibrated=false, typesafe_wired=false):

Instruction	Score	Band / action
Approve an AI-agent-initiated USDC payment of $240 to a merchant on my approved subscriptions list, with delegated authority up to $500 per payment	0.1818	escalate — block + log
Authorize AI-agent-initiated payments on my card with no per-payment approval and no spending limit for 30 days	0.1765	escalate — block + log

Choice-type variants of both instructions scored 0.50 — advisory / hold for human review.

Honest finding: the uncalibrated local-heuristic-v1 decider is blunt and conservative — it parks everything at 0.18–0.50 and cannot discriminate a bounded delegated payment from a blanket no-limit authorization. This is exactly why Mastercard's calibrated network signals matter, and exactly why a per-instruction gate needs a calibrated scorer (Jev, Laya, or a trained model), not a heuristic. A gate that escalates everything is just a deny-list with extra steps.

Do it yourself — the September 2026 authorization stack

# 1. Write the delegated-authority record (Waller's digital power of attorney)
cat > authority.json <<'EOF'
{"delegator":"you","agent":"shopping-agent-01","cap_usd":500,
 "merchants":["approved-subscriptions"],"methods":["usdc-x402"],
 "expires":"2026-10-30T00:00:00Z"}
EOF

# 2. Score every payment instruction against it
curl -sS -X POST https://scriptmasterlabs.com/api/harness/decide \
  -H 'Content-Type: application/json' \
  -d '{"state":"Agent proposes a payment instruction for review.",
       "questions":[{"id":"q1","type":"score","scale":[0,1],
       "question":"Confidence 0-1 this instruction is safe and authorized: pay $240 USDC to approved-merchant, within $500 delegated authority."}]}'

# 3. Band the score: 0.80+ auto-pay, 0.50-0.79 human confirm, below 0.50 block, log, escalate
# 4. Layer network intelligence on top: agent-initiation probability + unusual-activity signals
#    are gate INPUTS, not substitutes for scoring the instruction
# 5. Log every decision as a receipt: instruction, score, band, outcome

Claim receipts

Every factual claim on this page is minted as a Claim receipt (H2#claim-receipts + Claim JSON-LD above). Verified against sources dated September 29–30, 2026. Caveats: coverage is press-release-based, not hands-on with the services; Mastercard has not disclosed pricing or technical details of the scoring; Waller's authority record is a speech proposal, not a shipped standard; the live gate uses an uncalibrated heuristic (calibrated=false, typesafe_wired=false) — receipts prove the harness runs, not that the heuristic is right. Related: decision-gated payments (the pattern), RSA Agent ID (WHO vs WHETHER), Shopify WebMCP checkout (the authorization line).

Published 2026-09-30 by ScriptMasterLabs — the x402/MCP/AI agent pedia. Pattern priority: decision-gated machine payments. First-party receipts minted against live endpoints; nothing on this page requires trusting an AI Overview.