Autonomous transactions are financial or commercial transactions initiated and completed by software agents on a person’s behalf, often without real-time human involvement at checkout. An agent may search for an option, select a product or service, apply user-defined rules, and initiate payment within delegated limits.
This model changes the trust question. A financial institution or merchant must assess more than whether a person can access an account. It must also establish who authorized the agent, what the agent is allowed to do, whether the action remains within scope, and what evidence exists if the transaction is disputed.
Autonomous transactions are transactions that an automated software agent executes for a user after receiving permission to act within defined conditions. The user may set a broad objective, such as purchasing a flight below a spending limit, paying a recurring bill, or reordering a product when stock becomes available.
The agent can then perform several steps that a person would previously complete manually. These may include comparing offers, confirming availability, selecting a payment credential, submitting an order, and receiving confirmation. The human may approve the instruction in advance rather than interact with every individual payment.
Autonomous transactions are part of the broader idea of agentic commerce. Agentic commerce describes commercial activity in which AI agents interact with merchants, payment providers, and other services to complete tasks for consumers or organizations.
An autonomous transaction normally combines five elements: a user instruction, an authenticated agent, delegated authority, a payment or account credential, and controls that evaluate the transaction before completion.
These steps may occur across several systems. A merchant may recognize the agent, a payment scheme may attest to the consumer or credential, and an issuer may decide whether the payment should proceed. Interoperability matters because no single participant controls the whole transaction lifecycle.
Authentication verifies an identity or entity. Authorization defines the actions that identity or entity is permitted to perform. In an autonomous transaction, both questions apply to the human and the agent.
| Control | Question it answers | Example |
|---|---|---|
| Human authentication | Who approved the delegation? | Did the account holder authenticate with a phishing-resistant method? |
| Agent authentication | What software is making the request? | Does the request carry verifiable agent credentials? |
| Authorization | What may the agent do? | Can it spend up to a defined amount with approved merchants? |
| Transaction decisioning | Should this action proceed now? | Does the behavior match the instruction, context, and risk level? |
Authentication alone cannot establish that a transaction is legitimate. An authenticated agent could still operate outside its mandate, be compromised, or act on an instruction that has been changed. A trusted agent therefore needs verifiable delegation and ongoing authorization checks.
Autonomous transactions shift the main control point away from a visible checkout interaction, creating new opportunities for agentic commerce fraud. Fraud risk can arise when an instruction is created, when authority is delegated, when an agent accesses an account, when a payment is initiated, or after the transaction has been completed.
These risks make transaction evidence essential. A useful record should show the user’s instruction, the agent identity, the permission scope, the relevant payment credential, the decision signals, and any intervention or approval. Entersekt explores the wider banking risks of agent adoption in When bots break the rules: How banks can safely harness AI for growth.
Secure autonomous transactions require layered controls rather than a single approval event. The control set should cover delegation, identity, permissions, transaction context, monitoring, and recovery.
Authorize a clear action or policy through a trusted, phishing-resistant interaction. Verifiable intent links the authenticated instruction to what the agent may do, preserving defined limits and auditable evidence.
Distinguish approved agents from anonymous automation using signed requests and trusted key directories. Timestamps, expiry values, and nonces help establish origin and reduce replay risk.
Define what the agent may purchase, from whom, for how much, using which payment method, and for how long. Narrow permissions limit the impact of compromise or actions outside the approved scope.
Assess device, behavior, account history, payment details, location, velocity, and alignment with the original instruction. Authentication Advisor illustrates how real-time context can guide the authentication decision.
Entersekt’s article on AI fraud threats and how financial criminals exploit emerging technology explores the role of behavioral analytics and adaptive risk scoring.
Pause high-risk or ambiguous actions for fresh approval, stronger authentication, delayed payment, or review. Reserve intervention for situations where risk or uncertainty justifies it.
Banks and payment providers will need to extend fraud, authentication, and dispute processes across the full agent lifecycle. This includes customer enrollment, delegation, agent access, payment initiation, transaction monitoring, and post-transaction investigation.
For digital banking, the same principle applies to transfers and account changes as to purchases. An agent might schedule a payment, add a beneficiary, change a limit, or move money between accounts. These actions require controls that validate both the authority behind the request and the context in which it occurs.
Entersekt’s Context Aware Authentication applies risk, channel, and user context to authentication decisions across transactions. That approach is relevant to autonomous activity because the appropriate control can depend on what the agent is doing, not only on how it connected. This broader move toward decisions that connect identity, intent, risk, and evidence is explored in Four shifts redefining trust in digital payments.
Industry standards work is developing around three connected needs: trusted agent identity, verifiable user intent, and secure payment execution. The objective is to create interoperable signals that merchants, payment schemes, issuers, and agents can understand across organizational boundaries.
EMVCo’s work on agentic payments examines how consumer-authorized intent can be registered, referenced, retrieved, and managed across the transaction lifecycle. The FIDO Alliance is developing standards for agent authentication, verifiable user instructions, and trusted delegation for commerce. Visa’s Trusted Agent Protocol describes signed agent recognition, consumer recognition, and payment data that can help merchants verify requests.
These initiatives are developing areas rather than a single settled global standard. Financial institutions should therefore track the relevant payment scheme, identity, security, and regulatory requirements before committing to a production architecture.
Financial institutions should start with governance and control boundaries, then decide where autonomous execution creates measurable customer value. A practical assessment should cover the following questions:
A controlled rollout can begin with low-risk, narrow use cases and explicit limits. Monitoring should test not only payment outcomes but also delegation quality, unauthorized behavior, customer understanding, intervention rates, and the ability to revoke access quickly.
Autonomous transactions are broader than automatic payments because an agent can make decisions within delegated rules, while an automatic payment usually follows a predefined schedule and amount. Entersekt’s contextual authentication approach can help institutions assess higher-risk actions when an automated process changes behavior or transaction context.
An AI agent can complete a payment without real-time user involvement when the user has previously granted valid authority within defined limits. Entersekt helps financial institutions apply risk-aware authentication to sensitive transactions, so delegated execution remains subject to appropriate controls.
Banks can combine authenticated user delegation, agent credentials, signed requests, permission checks, and transaction monitoring to verify authority. Entersekt’s Authentication Advisor uses contextual risk intelligence to guide authentication decisions when an agent’s action requires additional assurance.
The response depends on the evidence, applicable rules, and the point at which the transaction was compromised. Institutions need records linking the user, instruction, agent, permission scope, authentication events, risk decisions, and payment outcome so they can investigate and assign responsibility.
Autonomous transactions are developing, but broad adoption depends on interoperable standards, clear liability rules, reliable agent identity, verifiable intent, and effective fraud controls. Entersekt’s focus on transaction authentication and contextual risk decisioning is relevant as institutions prepare for this shift.