Agent Transaction Probability Score.
An agent transaction probability score is a per-payment score, computed at execution time, estimating whether an AI agent truly initiated a transaction on the customer's authority and whether it fits expected behavior.
The term describes a category, not a standard. Nobody has published a definition of "agent transaction probability score," so treat this entry as a working definition. What makes the category distinct is timing and question. The score runs per transaction, at execution, and it asks about authority rather than criminality.
Mastercard's Probability Score is the first named instance. Coverage places it inside a four-layer stack: Know Your Agent identity verification, transaction-level probability scoring, payment execution with controls, and consumer protection signals. Cloudflare supplies web and payment signals and Skyfire supplies KYA verification. Thresholds, APIs, and pricing were not disclosed at announcement, so do not design against specific values yet.
For identity teams, the useful lesson is upstream. A score can only judge "within the customer's authority" if that authority is written down somewhere. Delegation records, agent token scopes, and consent history are CIAM data. If they are missing or unlinked, any score has to guess. The Know Your Agent for CIAM guide covers how to structure them.
Common questions
What is an agent transaction probability score?
It is a score computed for a single payment at execution time. It estimates whether an AI agent really initiated the transaction on the customer's authority and whether the transaction fits expected behavior. There is no standard definition of the term yet.
What is Mastercard's Probability Score?
Mastercard's Probability Score is part of its Agentic Commerce Trust Framework, announced September 30, 2026. It evaluates at execution time whether an AI agent initiated a transaction and whether it falls within expected behavior. It uses signals from Cloudflare and Skyfire and was in US-only testing at launch.
How is this different from a fraud score?
A fraud score looks for signs that a transaction is criminal. An agent transaction probability score asks whether the agent had the customer's authority and acted within expected behavior. A payment can be non-fraudulent and still fall outside what the customer delegated.
What does a CIAM team need to supply for agent transaction scoring?
Scoring depends on identity evidence: who the customer is, which agent they authorized, what limits they set, and when. A CIAM system that records delegations and agent tokens as linked, timestamped records gives a scoring engine something reliable to check against.
Related terms
In the guides
AI Fraud Detection in CIAM: Signup, Login, and Account Changes
Where fraud hits customer identity, which signals AI models score, and how to wire a fraud engine into signup, login, and account-change flows without wrecking conversion.
Authorization Patterns for Agentic Workflows: Delegation, Constraints, and Just-in-Time Permissions
AI agents need authorization models that handle delegated permissions, multi-step workflows, and least-privilege at machine speed. The patterns that work and the ones being invented.
Know Your Agent: Identity for AI Agents That Buy on a Customer's Behalf
What a CIAM team must build when a customer's AI agent shows up at login, account, or checkout: delegated consent, agent sessions, step-up, audit, and revocation.