Know Your Human: Sardine's Fix for the False-Decline Problem in Agentic Payments
TL;DR
Sardine launched its Know Your Human (KYH) protocol on 6 November 2026 to help merchants and issuers connect AI shopping agent transactions to the people behind them. Agent checkouts can look like bot traffic because they often come from cloud browsers, data-centre IP addresses and automation-style user agents. KYH passes human-linked signals (such as device ID, device fingerprint, IP address, email and phone) into the transaction context. It also supports an SMS verification step when a merchant needs more confidence. The strategic opportunity is to reduce false declines and restore evidence that can matter in chargeback disputes, without treating every agent as either automatically trustworthy or automatically suspicious.
Why can AI agent shopping create false declines?
AI agent purchases can be declined because conventional fraud systems may interpret a cloud-browser session as automated bot traffic, even when a real customer initiated it.
When a shopper uses an AI agent, the checkout may not come directly from the customer’s device in the way existing risk systems expect. Many agent shopping sessions run in cloud browsers, creating data-centre IP addresses and automation-style user-agent strings. Those signals can resemble malicious automation.
This creates a mismatch: fraud controls look for familiar signs of a human-operated session, but the customer is delegating the browsing and checkout activity. A legitimate purchase can therefore be caught by rules designed to stop bots. For payment operators, false declines are not only an acceptance issue; they can interrupt a customer journey before the merchant can distinguish authorised delegation from suspicious activity.
The broader challenge is that agentic commerce payments introduce a new identity question. It is not enough to detect that an agent is present. A merchant needs a way to connect that agent session with the human who initiated or authorised it.
Reliability is the other half of the equation: Visa’s Agentic Commerce Benchmark found that AI shoppers fail 42.8% of shopping runs and still complete a purchase in more than half of those failed runs, which raises the stakes on getting the right person attached to the right transaction. Our analysis of the benchmark results covers that reliability gap in detail.
How does Sardine’s Know Your Human protocol reconnect an agent to a person?
KYH makes a set of human-linked identifiers available to merchants and issuers when an AI shopping agent places a purchase.
The identifiers include device ID, device fingerprint, IP address, email and phone. Sardine says KYC and other identity details sit in extended use cases. The point is to add human context to an agent transaction, rather than relying only on signals generated by the agent’s cloud browser.
KYH can be implemented by agent developers or by merchants. Sardine describes two binding methods:
Agent-side SDK: The agent embeds Sardine’s SDK, which captures a device fingerprint and ties it to a session key.
Merchant Method URL: The agent requests a Method URL from the merchant and loads it in the background on the customer’s device, allowing the merchant to connect a device fingerprint to the agent session.
Human identifiers are passed through a small number of additional HTTP headers. In practical terms, the protocol aims to preserve a connection between the customer’s device context and the agent’s checkout session, even when the agent is operating through a cloud browser.
How does KYH step-up verification work?
KYH allows merchants to request an additional customer check when an agent checkout needs more verification, rather than blocking agent traffic outright.
When Sardine Device Intelligence detects checkout through a recognised AI agent (including Muse, Dot or Instinct) the merchant can send the customer an SMS containing a verification link. Opening the link connects the customer’s browser fingerprint with the fingerprint associated with the agent’s cloud browser. Each link can be opened up to three times and expires after five minutes.
This creates a pragmatic middle path for payment fraud prevention: an agent transaction need not be accepted without scrutiny, but neither must it be rejected solely because its session looks automated. Merchants can use a step-up action to seek a stronger link to the person behind the purchase.
The pattern will be familiar to operators who already use additional authentication for transactions that need more assurance. Here, the challenge is specifically to connect the human’s device context to the agent session. That is different from treating the agent itself as the customer.
Why does agent identity matter for chargeback disputes?
Agent activity can hide the device and IP signals that merchants rely on as evidence, creating a gap between the transaction and the customer in dispute processes.
That gap matters because existing dispute frameworks can depend on continuity in device or network data. Visa’s VAMP programme counts both fraud reports and disputes towards a merchant’s ratio. As a result, an agent-unauthorised chargeback can affect a merchant through both measures. Visa Compelling Evidence 3.0 requires the same device ID, fingerprint or IP address in two prior transactions within 120–365 days, signals that an agent’s cloud environment may mask. Mastercard’s First Party Trust also requires device and IP data.
KYH’s potential contribution is to help restore a chain linking the agent session to the human’s device and identity context. That does not guarantee that a merchant will meet every programme requirement or win a dispute; it addresses a specific evidence problem created when agent traffic obscures familiar signals.
For more on how card-not-present evidence requirements are changing, see our analysis of Visa’s chargeback rules and Compelling Evidence 3.0.
Where does KYH fit in the agentic payments stack?
KYH is an identity layer that complements agentic commerce protocols, rather than competing with the protocols that govern authorisation and payment.
That distinction is important for payment providers. Knowing that an agent is operating, or understanding the payment flow it uses, does not by itself prove which human initiated the session. KYH is aimed at that missing relationship: linking the human to the agent’s activity.
For POS companies, PayFacs, ISOs, issuers and other payment operators, the immediate operational question is whether their fraud stacks can recognise agent traffic at all. If agent sessions are consistently classified as bots, legitimate use may be declined before other controls can assess the transaction. If systems cannot capture or use human-linked signals, merchants may also lack useful context when a payment is challenged.
As the wider protocol landscape develops, it is useful to distinguish agent identity from payment execution. Our overview of Visa’s first live agent protocol pilot covers a separate part of that evolving stack. KYH’s stated focus is the identity connection, and Sardine says it is complementary to existing agentic commerce protocols.
Sardine has implemented KYH in its own Personal Agent and is seeking design partners. The protocol is also extendable to KYC pass-through to banks for agentic banking use cases. For now, that makes it an approach for operators to assess, not a reason to assume every agent transaction will carry the same identity signals.
Frequently asked questions
What is Know Your Human?
Know Your Human is Sardine’s protocol for making human-linked identifiers available to merchants and issuers when an AI shopping agent places a purchase. Its purpose is to connect the agent’s transaction context to the person behind it.
Why do AI agent purchases get declined?
Agent purchases can come from cloud browsers using data-centre IP addresses and automation-style user-agent strings. Fraud systems may interpret those signals as bot activity and decline a legitimate customer’s transaction.
How does KYH verification work?
KYH can bind a customer’s device fingerprint to an agent session through an embedded Sardine SDK or a merchant Method URL loaded on the customer’s device. Merchants can also send an SMS verification link to connect the customer’s browser fingerprint with the agent’s cloud-browser fingerprint.
What does KYH mean for chargebacks and disputes?
KYH aims to help preserve a connection between the agent session and human-linked device or identity signals that may be relevant to dispute evidence. It does not, by itself, guarantee that a transaction meets a card programme’s evidence requirements or that a dispute will be resolved in the merchant’s favour.
Is KYH a standard or a product?
KYH is Sardine’s protocol, and the company is seeking design partners. It is described as complementary to existing agentic commerce protocols, not as a replacement for them.
Key takeaway
Sardine’s Know Your Human protocol targets a specific weakness in agentic commerce payments: the gap between the person authorising a purchase and the automated session that reaches checkout. By connecting human-linked identifiers to agent activity, and offering an SMS step-up when needed, KYH proposes a way to reduce false declines and improve the context available for disputes. Payments operators should assess whether their fraud systems can recognise agent traffic and preserve the evidence needed to make decisions about it.