Banks that move customers to synced passkeys usually settle the login question first: the passkey is phishing-resistant, user verification runs on every assertion and the customer no longer needs a password or an SMS code to get in. The harder question comes one step later. When the same customer confirms a payment to a new payee, is the passkey assertion on its own enough?

Banking Passkeys Report. Practical guidance, rollout patterns and KPIs for passkey programs.
For most banks the honest answer is that they want one more signal. A synced passkey lives in Apple, Google or a third-party credential manager and follows the customer to every device that joins that account. That is exactly what makes it usable at scale, and it is also why it cannot tell the bank which device is in front of it. The supplementalPubKeys extension that was meant to close this gap in WebAuthn was deprecated in 2024, so relying parties have to build the trusted device signal themselves.
This article covers:
A common mistake is to make one decision for "passkeys" and apply it to every journey. Either the risk team accepts synced passkeys everywhere, including high-value payments, or it rejects them everywhere and keeps SMS OTP for login as well. Both outcomes are worse than splitting the decision by what each journey has to prove.
A successful assertion from a synced passkey proves three things: the request came from the bank's own origin (so it cannot be phished on a look-alike domain), the customer holds the private key and the authenticator performed user verification with biometrics or a device PIN.
It does not prove that the device is one the bank has seen before. Apple describes how passkeys reach new devices either by pairing with an existing device or through iCloud Keychain recovery, and Google Password Manager and third-party managers follow comparable models. Whoever controls the credential manager account and its recovery factors can therefore bring the passkey to a new device. That is the account recovery path attackers target, as covered in detail in synced passkey attacks.
NIST takes the same view in its final SP 800-63B-4 guidance on syncable authenticators: syncable authenticators may be used up to AAL2, and agencies should assume that they may be shared. For login that is an acceptable trade-off, as covered in our NIST passkeys analysis. For moving money, most risk teams want a device signal on top.
Under PSD2, Article 7 of the Regulatory Technical Standards on strong customer authentication asks payment service providers to mitigate the risk that possession elements are used by unauthorised parties and to prevent their replication. This applies to every strong customer authentication, for account access as well as for payments. Corbado's view is that synced passkeys can meet SCA although they are not device-bound. Many European banks still take the conservative reading, classify synced passkeys as one factor and make the 1FA vs 2FA decision explicitly. Banks that classify them as one factor need an additional factor on unknown devices at login too, unless an exemption applies.
Independent of the legal classification, a known-device check is a strong risk control for the moments when money leaves the account. Outside the EU the legal wording differs, but scam-driven regulation in markets like Australia and the UK pushes in the same direction: stronger checks on payments to new payees and other high-risk actions.
The following diagram shows the resulting layered model for banks that accept synced passkeys for login: the passkey stays the same and the additional signal depends on the action. Banks that classify synced passkeys as one factor apply the step-up on unknown devices at login as well.
Low-risk actions such as payments to known payees rely on the bank's existing risk signals, while the known-device check is reserved for high-value payments, new payees and limit changes. This keeps the passwordless experience for the vast majority of sessions and concentrates friction where the harm is, the same principle behind step-up authentication and adaptive MFA.
See how many people actually use passkeys.
Once the bank decides that payments need a known-device check, the next question is which signal to use. On the web there are five realistic options.
| Signal | What it proves | Strengths | Weaknesses |
|---|---|---|---|
| Server-issued first-party cookie | This browser was trusted before | Works in every browser, no extra hardware, easy to revoke | Cleared by users and browsers, can be copied by infostealer malware |
| Device Bound Session Credentials | The request comes from the machine that holds a non-exportable key | Stolen session cookies expire within minutes and cannot be refreshed from another machine | Chrome 146+ on Windows, macOS rolling out, listed in Edge 147, no Safari, Firefox or mobile yet |
| Native banking app (app push or in-app key) | The customer's registered app approved the action | Strongest signal most banks already have | Needs the app installed, adds a device switch on desktop |
| Device fingerprinting | The browser looks similar to one seen before | Invisible to the customer | Probabilistic, weakened by privacy features, privacy and consent questions |
| Device-bound passkey or security key | A specific hardware authenticator was used | Real possession proof | Extra enrollment, lost-device recovery, cost for issued keys |
Most banks end up with a combination: the native app establishes trust once, a first-party cookie carries that trust on the web, a DBSC-bound session adds hardware binding where the browser supports it and fingerprinting stays a risk-engine input only. Device fingerprinting has clear limits against modern infostealers, which is why it should not be the only signal for payments.
Google announced public availability of DBSC for Windows in Chrome 146 in April 2026, with macOS following, and Edge lists it from version 147. DBSC protects short-lived session cookies that the browser refreshes by proving possession of a key in the TPM or Secure Enclave. Where it is available, tie the trusted-device record to the DBSC session so that trust is re-proven on every refresh. Today DBSC covers only part of a bank's customer base, so the long-lived first-party cookie remains the baseline.
The cookie looks simple, but most of the design work sits in four questions: how trust is granted, what the cookie proves, how it behaves over time and what happens without it.
The cookie is only as strong as the step that issues it. A passkey login alone cannot grant device trust, because that would turn the passkey back into the only factor. Typical trust-granting steps are:
Some banks also mark devices as trusted after SMS OTP or based on fingerprinting. That is the weakest option for payments, because the gap the cookie closes is exactly the one that SIM swap and social engineering exploit. If a bank uses it, it should limit what a cookie issued that way unlocks.
There are also moments when trust should not be granted automatically:
The cookie should carry only an opaque reference. The trust decision stays on the server:
HttpOnly against script access, Secure and the __Host-
prefix
so the cookie is host-only with Path=/ and no Domain. Prefer SameSite=Lax: with
Strict the cookie is missing on the first request after a cross-site redirect, for
example from a merchant site into the confirmation page. Embedded or POST-based flows
such as 3-D Secure challenges need separate handling.When confirming a high-value action, the bank requires both: a fresh passkey assertion and a valid cookie whose record matches the customer and, ideally, the credential that just signed. If the cookie is missing or its record does not match, the payment steps up.
This is where cookie-based designs fail in production, usually silently.
Set-Cookie header from the bank's own server on the bank's own
domain.Decide this case explicitly, because it will be common. For banks that classify synced passkeys as meeting SCA for login, the passkey login still succeeds and the customer is never locked out because a cookie is gone. Only actions that need device assurance trigger the step-up, with a short explanation ("You're using this browser for the first time for payments. Please confirm in the app.") and a fallback for customers without the app. The step-up then issues a fresh cookie, so the next payment on that browser is smooth again.
Try passkeys in a live demo.
The most useful advice for teams at this stage: start setting the cookie before anything depends on it. In shadow mode the bank issues and reads the trusted-device cookie, logs what the policy would have done and changes nothing for the customer. Login and authorisation behave exactly as before. The following timeline shows the three phases.
Shadow mode answers the questions a risk committee will ask before it approves enforcement:
Define thresholds before enforcing, for example a minimum survival rate on the top platforms and a maximum step-up share, and keep monitoring after go-live. A browser update that changes storage behavior shows up as a sudden drop in cookie survival on one platform. The server sees that the cookie is missing. Client-side context such as browser version, private mode and storage behavior helps explain why. Without it, the team learns about the cause from step-ups and complaints a week later, which is the kind of blind rollout mistake banks try to avoid.
Igor Gjorgjioski
Senior Product Lead, VicRoads
We hit 80% mobile passkey activation across 5M+ users without replacing our IDP.
See how VicRoads scaled passkeys to 5M+ users, alongside their existing IDP.
Read the case studyOnce a trusted-device record exists, it is tempting to use it for more. The most valuable second use is login convenience: a browser with a valid record can offer an enumeration-safe one-tap passkey button that requests the known credential directly instead of relying on autofill.
That is a good idea as long as the two meanings stay separate in the record:
Teams that merge both into one flag end up either weakening payments (every login creates assurance) or degrading login (one-tap only appears after an app confirmation). A single cookie can carry both, as long as the server record stores the level and how it was reached.
Subscribe to our Passkeys Substack for the latest news.
Whether the cookie arrives is visible on the server. Why it disappeared on a given OS and browser and what happened in the passkey prompt when a payment was confirmed never reach a server log. Corbado Observe captures that layer on top of the existing login and payment flows with one script, without touching the authentication itself.
See Login Funnel in Corbado Observe →Corbado Connect adds the flow itself for banks that want it: passkey login with enumeration-safe one-tap on known devices, on top of the existing identity stack.
Synced passkeys are a strong login factor, and banks should not hold back their rollout because of payment requirements. The payment question is real but separate: a synced passkey proves the customer, and the device needs its own signal. Answer it with a known-device signal for high-value actions only, built from a stronger trust-granting step, a well-designed first-party cookie and DBSC where browsers support it. Start that cookie in shadow mode during the passkey pilot, measure how it behaves on real devices and enforce it only once the numbers hold.
Synced passkeys prove that the customer controls the credential and passed user verification, which is strong enough for login. They do not prove which device is in use, because the private key can be restored on any device that joins the customer's credential manager account. For high-value transaction confirmation, banks add a known-device check on top of the passkey assertion.
The most widely available option on the web is a server-issued first-party cookie that references a server-side trusted-device record. Trust is established once through a stronger step such as confirmation in the banking app. Where available, Device Bound Session Credentials in Chromium browsers bind the session to a hardware key so stolen session cookies cannot be replayed from another machine.
For banks that classify synced passkeys as meeting SCA for login, the login should still succeed. Only actions that need device assurance, such as payments to new payees or limit changes, trigger a step-up that re-establishes trust on that browser. Treating a missing cookie as a login failure creates lockouts, because browsers clear cookies for many legitimate reasons.
Setting the cookie without letting it change any customer-visible behavior shows how long it survives per operating system and browser and how many transactions would have needed a step-up. Those numbers decide whether the signal is reliable enough to enforce and what step-up volume the contact center should expect.
Related Articles
Table of Contents