Agentic commerce

The vanishing point-and why agentic payments require a new layer of visibility

hero
  • Insight
  • October 30, 2026

The moment humans disappear from the transaction, new risks emerge. A new generation of verifiable credentials can enable trust and fulfill the promise of autonomous purchasing.


This four-part series was produced in collaboration with EQTY Lab.

Ilana Golbin Blumenfeld

Ilana Golbin Blumenfeld

Principal, Trust AI, PwC US

Kelly Guinan

Kelly Guinan

Director, PwC US

Contributors

Jonathan Dotan
Founder and CEO of EQTY Lab

Key takeaways:

  • Existing payment standards have gaps. Both traditional standards such as ISO 8583, ISO 20022, and 3DS and new agentic commerce protocols may not convey key signals, such as customer intent, in a uniform way to all stakeholders, or are in early stages of adoption among the various payment types.
  • Verifiable credentials are one potential path to shared trust for agentic signals. They can bind an agent's identity, authorization, and conduct evidence that travels with the transaction, so evidence reaches the parties who carry the responsibility.
  • KYA is the new KYC. Shared trust in an agentic transaction depends on a common “know-your-agent” model to establish evidence, accountability, and liability.
  • Context and responsibilities are split. Contextual information originates upstream, but responsibility stays downstream. Issuers, merchants, acquirers, networks, and customers should have a shared way to exchange and confirm new and enhanced agentic signals.

By 2030, PwC projects that global ecommerce will reach $73.5 trillion. AI is the dominant driver of that growth, with over 35% of generative AI (GenAI) traffic replacing search. Agents will represent one of the largest generational shifts in commerce, a shift already underway as consumers begin purchasing through the recent proliferation of personal AI agent assistants.

Yet, that opportunity comes with novel challenges. Agents introduce new operational, fraud, and dispute risks, driven by the different transaction patterns those produce. Issuers are already responding to this visibility gap. In October 2026, an industry-leading bank’s CEO said their bank will not allow outside AI agents to operate retail customer accounts unless through authorized connectors, citing security and liability risk. As consumer adoption increases, the financial services ecosystem will need to consistently answer basic questions about each transaction.

The payment systems of today don’t have a clear way to represent an agent to those managing the risk of the transaction. In this blog we describe this vanishing point and how you can make those missing signals visible.

Jonathan Dotan,Founder and CEO of EQTY Lab

These four questions set the foundation for addressing emerging risks in agentic commerce. Getting them right can help preserve the speed, trust, and consumer protections that have made card payments work.

The vanishing point: where human signals disappear and new risks emerge

Right now, as the credit card authorization message moves through the network rails from merchant to issuer, many of the data elements that would indicate the transaction was agent-initiated are not consistently carried forward by default. The same is true for additional context supporting the substance of the agentic transaction, as merchants and networks may hold information that never reaches the party making the authorization decision. This is the vanishing point.

Two types of signals tend to fade:

  • The traditional signals built and used to identify fraud based on a person’s device, behavior, and history (e.g., device fingerprinting, velocity checks, etc.) may fail to capture the agent’s origination upstream.
  • The new agentic signals that establish an agent's mandate and competence (e.g., machine-readable customer instructions, spending limit values, agent trust grade, etc.) may not flow downstream to the issuing banks.

Both gaps point to the same need: a way for context to reach the parties who may need to act on it.

Emerging protocols

Over the last 18 months, PwC has tracked over a dozen proposed protocols and software development kits (SDK) from payment networks, frontier labs, and hyperscalers to define how agents can prove user authorization, identify themselves, interact with merchants, and initiate payments. Issuers, networks, and major technology and AI companies have introduced new standards and governance initiatives with at least five new efforts announced in 2026 alone. Numerous initiatives have been announced throughout 2026 between networks and issuers to pilot agentic transactions to unearth and respond to many of these questions.

These are meaningful developments. They show a collective understanding that the agentic commerce ecosystem needs more than a standard payment credential. It should have agent identity, customer authorization, merchant verification, cryptographic proof, evidence references, and ways to distinguish legitimate agent traffic from fraud or bot activity. Each protocol is solving a genuinely hard problem, but they may be solving different problems without a clear standard as to how issuers might receive context shared by upstream parties. Even with this meaningful progress, the landscape for documented agentic signals across these agentic protocols remains fragmented, and forward-looking signals, such as runtime verification, are not yet in place.

Trust will be a key component to a future where agents can truly operate on our behalf autonomously. We need to prepare our payment processes to gather the information needed to enable that trust.

Ilana Golbin Blumenfeld,Principal, Trust AI, PwC US
Protocol What problem does it solve?
MCP Lets an agent connect to any tool or data source through one standard interface.
A2A Standardizes how an agent finds and negotiates with another agent.
ACP / UCP

ACP: Lets a chat assistant complete checkout directly in chat. 

UCP: Lets an agent read a merchant’s product catalog and pricing details.

AP2 / TAP

AP2: Defines a mandate that confirms an agent’s spend request matches customer authorization.

TAP: Verifies that a purchase request comes from a registered, authenticated agent.

Cards / x402

Cards: Moves money on existing card rails, scoped with agentic controls.

x402: Makes per-API-call payments in stablecoin.

While there has been some collaboration and convergence across these protocols, they do not yet add up to a shared operating model for a strong “know your agent” (KYA) framework that would let stakeholders confirm who initiated a transaction and under what authority. EMVCo, the global technical body behind card payment specifications, took an early step toward that model in September 2026 with the release of its draft Agentic Payments Framework. While the timing and shape of formal specifications are not yet disclosed, this is a significant attempt to move the needle on both standardizing KYA expectations and consistency in identifying agentic transactions across the ecosystem.

“Human-not-present” informing “know your agent”

Whatever form these frameworks take, they will draw upon lessons from historical "know your customer" models. What once represented a customer’s direct decision to buy now represents an agent’s interpretation of prior instruction. Networks have already defined one version of this in card-not-present transactions, distinguishing customer-initiated from merchant-initiated charges. Agentic commerce takes that separation a step further. Whereas the four questions above highlight whether a transaction can be trusted at authorization, what follows are the practical stakes for each participant when those questions go unanswered in a “human-not-present” transaction. Without a shared solution, the consequences fall differently on each participant:

It can create ambiguity if they are unable to identify if a transaction was agent-initiated. They may approve transactions that later become disputes or decline legitimate agentic purchases because these trigger human-centric fraud signals.

It can create evidentiary risk. They may accept agentic transactions without knowing what evidence they should preserve to defend against chargebacks. It can also create representment risk; without evidence that the agent acted within its mandate, a defensible transaction can become an unrecoverable loss.

The lack of a clear representment path may create higher chargeback rates within their merchant portfolios; acquirers may be exposed to additional penalties given chargeback-related scrutiny and monitoring from the networks.

It can create governance pressure. To help arbitrate human-not-present commerce, they should adapt reason codes, evidence standards, liability guidelines, and operating rules specifically built for agent-initiated transactions.

It creates accountability risk. Convenience breaks down if no one can explain who is responsible if an agent acts outside established expectations.

They are practical and essential challenges when determining whether any agentic transaction can be authorized, accepted, defended, and disputed with confidence at machine speed.

One promising solution: verifiable credentials

In May 2026, EMVCo published a draft Digital Payment Credential specification, endorsing verifiable credentials (VCs) for cryptographically securing agentic payment flows. This move aligns with a broader industry push from global industry coalitions such as the Coalition for Secure AI (CoSAI) to establish how autonomous agents should be treated as verifiable, governed identities.

VCs provide a digital proof of an agent’s identity and its context in a secure, tamper-evident format. What makes them suited to closing the gap between what upstream parties know and what downstream parties can act on is that they are:

Trust in an agent should be derived from always-available cryptographic evidence that can be checked at each decision point and trust boundary in a payment flow, rather than relying on a single operator's assurance. This allows gateways to reliably enforce rules ranging from agent identity to customer mandate to merchant verifications.

Signals should reach each party that needs to act on them—something that does not happen today. A portable credential allows any signal to travel with the transaction with verifiable provenance. This enables any party to confirm it, and coalition partners to share it without requiring bespoke integrations.

Payment systems require proof of long-term credentials that can be exchanged at any point, upon policy decision, for scoped tokens authorizing agents to take specific actions or access critical resources.

The industry's commitments to privacy-enhancing technologies (PETs) cannot be sacrificed. Selective disclosure, a principle EMVCo already builds on, should extend to the agentic context.

Assessing whether an agent acted within its authority requires visibility into its runtime state at the moment of execution—evidence that has to come from the compute environment itself, ideally anchored in hardware rather than asserted in software.

Credentials give the ecosystem a way to carry this evidence, and settling on that approach is one part of the work. The rest is agreement on what context travels with a transaction, what evidence should be retained, and how an agent's authority is confirmed over time.

Where do we go from here

The existing payment ecosystem is moving to adopt a practical version of “know your agent”. The questions posed sound simple but answering them across the transaction lifecycle requires changes to how context, evidence, assurance, and rules are handled.

For agentic commerce to be successful, the payment ecosystem should have:

Agentic transactions should have a consistent way to identify the agent, the platform or operator behind it, the mandate it was acting under, and the scope of its authority. The proper participants need just enough information to understand whether the transaction was agent-initiated and whether it appeared to be within bounds.

Customer instructions, mandate scope, revocation status, agent action logs, merchant verification records, and payment-enabler activity may need to be retained and retrievable when disputes arise.

Agent platforms should be identifiable, credentialed, monitored, and revokable when needed. Human-not-present commerce cannot rely only on trust established at onboarding. It should have ongoing assurance that the agent is still acting safely and within authority, and that the controls around the agent are effective. The role that agents themselves play in addressing risks and maintaining the security of the payment's ecosystem should be defined, consistent, and effective.

The promise of agentic commerce can only scale if the payment ecosystem can preserve trust after the vanishing point. Over the next several weeks, this series plans to take a deeper dive into the risks facing issuers and merchants navigating the emergence of agentic commerce while seizing the opportunity presented by it.

Contact us

Ilana Golbin Blumenfeld

Ilana Golbin Blumenfeld

Principal, Trust AI, PwC US

Kelly Guinan

Kelly Guinan

Director, PwC US

Follow us

Required fields are marked with an asterisk(*)

Your personal information will be handled in accordance with our Privacy Statement. You can update your communication preferences at any time by clicking the unsubscribe link in a PwC email or by submitting a request as outlined in our Privacy Statement.

Hide