---
url: 'https://www.corbado.com/blog/passkey-login-account-enumeration'
title: 'Passkey Login & Account Enumeration: Decision Guide'
description: 'Passkey login and account enumeration: what username-first leaks, what usernameless costs in passkey usage and how to decide.'
lang: 'en'
author: 'Vincent Delitz'
date: '2026-11-22T11:58:23.058Z'
lastModified: '2026-11-22T13:22:17.915Z'
keywords: 'passkey account enumeration, passkey login enumeration, allowCredentials privacy leak, one-tap passkey login, conditional ui account enumeration'
category: 'Passkeys Strategy'
---

# Passkey Login & Account Enumeration: Decision Guide

## Key Facts

- In a **username-first** passkey login, the `allowCredentials` list in the server response
  is the enumeration signal: it shows that an account exists and has a passkey, and its
  shape can hint at the authenticator type.
- **Passkey autofill** (conditional UI) is enumeration-safe on the server side because the
  browser lists passkeys locally and the server answers every visitor the same way.
- A generic passkey button alone is pressed by only around **10-25%** of passkey holders
  (aggregate Corbado observation). Without a returning-browser shortcut, a usernameless
  login can lose up to half of the passkey logins a username-first login would get.
- A **one-tap passkey button**, offered to browsers with a first-party cookie from an
  earlier passkey login, closes most of that gap in both login models.
- The value of the enumeration signal and the cost of going usernameless both shrink as
  phishable fallbacks disappear, so the right answer can change over the rollout.

## 1. Introduction

Sooner or later, many passkey projects at banks, retailers, public services and platforms
with a strict security team hit the same wall. The product team wants the passkey dialog to
open as soon as the user types their username, because that gives the highest passkey
usage. The security team has a long-standing rule against account enumeration: an
unauthenticated visitor
must never learn whether an account exists. A username-first passkey login breaks that rule
in a new way, because the step after the username now looks different for users with a
passkey.

Without numbers, the discussion stalls. Security says "no signal", product says "no
adoption without the automatic prompt", and nobody can say what each side gives up. This
guide provides the material for a decision:

- **What can a passkey login reveal, and where does the leak sit?**
- **Which entry points leak and which do not?**
- **How many passkey logins does an enumeration-safe design cost?**
- **What makes a username-first signal small, and what do decoys achieve?**
- **How does the trade-off change between launch and a passkey mandate?**

## 2. What a Passkey Login can reveal

[OWASP's Authentication Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Authentication_Cheat_Sheet.html)
sets the goal: login, recovery and registration should respond in a generic way, with the
same message, the same HTTP status and the same processing time, whether the account exists
or not. With passwords, the login step met that goal easily, because the password form
looked the same for everyone. Passkeys add a step that depends on the account.

### 2.1 Two new Account Facts

Login pages have always risked leaking that an account exists, that it is locked or where
one-time codes are sent ("Code sent to j•••@gmail.com"). Most security teams already have a
position on these. A username-first passkey login adds two facts that need a fresh
decision:

- **The account has a passkey:** Only accounts with a passkey get WebAuthn request options
  after the username. Everyone else gets the password screen.
- **A hint at the passkey provider:** The number, length and transports of the credential
  IDs in the response can hint at what kind of authenticator created them.

![Five account facts a username-first login can reveal, with the two new passkey facts and how each one leaks](https://s3.eu-central-1.amazonaws.com/corbado-cloud-staging-website-assets/passkey_login_enumeration_account_facts_0fa51f489e.png)

### 2.2 allowCredentials List is the Signal

In a username-first flow, the browser sends the username, the server looks up the account
and returns WebAuthn request options. For an account with a passkey, those options contain
the [allowCredentials](https://www.corbado.com/glossary/allowcredentials) list, so the operating system shows only
this account's passkey:

```json
{
  "challenge": "9fQ…",
  "rpId": "example.com",
  "allowCredentials": [
    { "type": "public-key", "id": "Kx3q…9A", "transports": ["internal", "hybrid"] }
  ]
}
```

For an account without a passkey or an unknown username, the server returns no options and
the browser moves on to the password screen. An outsider who repeats the first request with
a list of usernames can sort them into "has a passkey" and "does not". The W3C describes
this in
[WebAuthn Level 3, section 14.6.2 Username Enumeration](https://www.w3.org/TR/webauthn-3/#sctn-username-enumeration).

[Section 14.6.3 Privacy leak via credential IDs](https://www.w3.org/TR/webauthn-3/#sctn-credential-id-privacy-leak)
adds a second point: the length of a credential ID "might be a hint as to what type of
authenticator created it", and the number and lengths of IDs might work as a correlation
handle across sites. Combined with the `transports` values, this can help a phishing message
sound credible. The same section names the cleanest prevention: use discoverable
credentials, so the `allowCredentials` list is not needed at all.

### 2.3 Why the Signal matters while Fallbacks are phishable

The passkey itself cannot be phished. Knowing that an account has a passkey gives an
attacker nothing against the passkey. The signal matters because of what surrounds it:

- **Targeted phishing:** A list of confirmed accounts makes spear phishing cheaper and more
  convincing.
- **Phishable fallbacks:** As long as password, SMS OTP and recovery flows remain open, the
  attacker uses the confirmed account to attack the fallback, which is the path that can
  actually be phished.
- **Privacy expectations:** For some services, membership alone is sensitive. Knowing that
  a person has an account at a health service, a debt advice service or a dating platform
  is a privacy issue independent of account takeover.

Many security teams therefore reject a visible signal during the phase when most users
still log in with a password. Their concern is the fallback that follows the signal.

## 3. Entry Points and what each reveals

A [passkey login](https://www.corbado.com/blog/passkey-login-best-practices) starts either with the username or
with the passkey. In practice, login pages combine several entry points, and each one has
its own enumeration profile.

| Entry point                                  | Who sees it                          | Reveals account state?                        | Main limitation                                          |
| -------------------------------------------- | ------------------------------------ | --------------------------------------------- | -------------------------------------------------------- |
| Typed username, then automatic passkey prompt | Everyone                             | Yes, through `allowCredentials`               | Prompt on a new device ends in a QR code                 |
| Typed username, empty `allowCredentials`      | Everyone                             | No                                            | Dialog lists every passkey for the site or a QR code     |
| Passkey autofill (conditional UI)            | Users with a passkey on this device  | No                                            | Password autofill competes in the same field             |
| Generic passkey button                       | Everyone                             | No                                            | Only around 10-25% of passkey holders press it           |
| One-tap passkey button                       | Browsers with an earlier passkey login | No (only to that browser)                     | Gone after cookies are cleared or on a new browser       |

Three observations matter for the decision:

- **Only the automatic prompt after a typed username leaks.** Every other entry point
  behaves the same for every visitor who names an account.
- **An empty list after the username is safe but pointless.** It keeps the username step
  and loses its benefit: the dialog no longer shows only this account's passkey. If you go
  enumeration-safe, go usernameless properly.
- **Conditional UI is safe from the server's view.** It starts a background WebAuthn request
  with an empty `allowCredentials` list and the browser shows matching passkeys locally.
  Some security teams still object to any page that lists passkeys before the user has
  typed an identifier, mainly on shared devices. Clarify that position early, because it
  decides whether [conditional UI](https://www.corbado.com/blog/webauthn-conditional-ui-passkeys-autofill) is
  available to you at all.

For the username-first prompt, the start rule matters too. Prompting every account with a
passkey sends users on a new laptop into a QR code they do not expect. Good start rules only
prompt where the passkey is likely available on this device, for example based on a cookie
from an earlier passkey use or
[passkey intelligence](https://docs.corbado.com/corbado-connect/features/passkey-intelligence).

## 4. What Usernameless costs in Passkey Logins

Security teams can name the risk. Product teams need a number for the cost. The ranges
below come from Corbado's aggregate observations across passkey rollouts and describe
logins by users who already have a passkey. Your values depend on platform mix, audience
and login frequency, so treat them as a starting hypothesis to validate.

### 4.1 Size of the Usage Gap

Compared against a username-first login with a one-tap passkey button:

- **Usernameless with one-tap:** around 10-20% fewer passkey logins. Users who type their
  username go to the password, because nothing may react to the typed account.
- **Usernameless without one-tap:** up to half fewer passkey logins. Only passkey autofill
  and a generic button remain, and only around 10-25% of passkey holders press a generic
  button.
- **Username-first without one-tap:** also loses passkey logins, because the prompt depends
  on the user typing a username that leads to a passkey on this device.

The press rate needs context. It is measured among users who already have a passkey. The
passkey login rate across all logins, which earlier Corbado material puts at around 5% for
button-only designs, is much lower because most users have no passkey yet. Both numbers
point the same way: a button alone does not change login habits.

Two conclusions follow:

- **One-tap is the biggest single lever** in both login models.
- **The login model changes usage and leaves enrollment untouched.** It does not change how
  many users can create a passkey or do create one.

### 4.2 Same Enrollment, different Passkey Login Rate

The share of all logins that use a passkey is the product of four factors: device support,
availability of the passkey at login, enrollment and usage. The following illustrative
comparison keeps the first three factors identical and changes only usage.

The values are illustrative. The pattern holds generally: an enumeration-safe design leaves
enrollment untouched and can lower the passkey login rate by up to a factor of almost two,
depending on how well the returning-browser shortcut works. Adoption forecasts built on
username-first usage need to be recalculated whenever an enumeration-safe design is on the
table.

### 4.3 One-Tap depends on Cookie Durability

A [one-tap passkey button](https://docs.corbado.com/corbado-connect/features/one-tap-login)
is enumeration-safe because it never reacts to a username the visitor types. After a
successful passkey creation or login, the server sets a first-party cookie that records
prior passkey use in this browser. On the next visit, the page offers a one-tap button for
that account. A visitor without the cookie sees the normal page. For the shared-device side
of remembered identifiers, see [Remember Me & Passkeys](https://www.corbado.com/blog/remember-me-passkeys).

How much one-tap recovers depends on how many users return while the cookie still exists:

- **Set it server-side on your own origin:** WebKit's
  [Intelligent Tracking Prevention 2.1](https://webkit.org/blog/8613/intelligent-tracking-prevention-2-1/)
  caps persistent cookies created through `document.cookie` at seven days. Set the marker in
  an HTTP response as `Secure` and `HttpOnly`, served from your own infrastructure and not
  from a vendor subdomain behind a CNAME, where Safari can apply similar caps.
- **Respect the browser ceiling:** Chrome
  [limits cookie lifetimes to 400 days](https://developer.chrome.com/blog/cookie-max-age-expires)
  since Chrome 104. Longer expiry values are silently shortened.
- **Expect clearing:** Users clear cookies, use private windows or switch browsers. On a new
  browser, one-tap only appears after the first passkey login there.
- **Expect stale markers:** The server cannot know whether the private key still exists on
  the device. If the user deleted the passkey, the one-tap attempt ends in a QR code. Offer
  a clear route back to the normal login. When a credential is unknown, use
  `signalUnknownCredential` from the [WebAuthn Signal API](https://www.corbado.com/blog/webauthn-signal-api), which
  the spec recommends over `signalAllAcceptedCredentials` toward unauthenticated callers.

To estimate coverage, combine two distributions: the time between logins per user from your
login logs and the practical cookie survival per browser and operating system. A service
whose users log in weekly keeps most of the one-tap benefit. A service whose users log in
twice a year loses a large part of it, and its usernameless gap moves toward the "without
one-tap" case.

## 5. Option A: Accept the Signal and keep it small

An organization that accepts the username-first signal for higher passkey usage should
manage it actively. The controls fall into three groups.

![Seven controls that keep a username-first enumeration signal small, with an effort rating for each](https://s3.eu-central-1.amazonaws.com/corbado-cloud-staging-website-assets/passkey_login_enumeration_option_a_controls_76bf2951bf.png)

**Make responses indistinguishable where possible:**

- Unknown usernames get the same password step as accounts without a passkey.
- The account lookup runs on every path, with the same timing for unknown usernames.

**Make harvesting slow and visible:**

- Rate limits per IP address, per username and overall.
- Bot protection at the edge and a device and IP reputation check before the lookup.
- Alerts on spikes in username lookups.

**Protect what an attacker would do next:**

- Extra checks on SMS codes and [recovery](https://www.corbado.com/blog/passkey-fallback-recovery) for accounts
  with a passkey.
- Clear communication that the service never asks for one-time codes.

The first two groups make enumeration slower and more expensive. They do not remove the
disclosure. A patient attacker who tests a few hundred targeted usernames from distributed
residential IPs stays below most rate limits, so existing bot and fraud tooling should be
tested against targeted, automated and distributed enumeration attempts before anyone
relies on it.

The third group has the most effect. The signal only helps an attacker who can follow it
with a phishable step, so hardening the fallback protects the path an attacker would
actually use.

## 6. Option B: Decoy Responses

WebAuthn Level 3 offers a middle path in section 14.6.2: for an unknown username, continue
the ceremony with "plausible imaginary values". If every username, known or not, receives
passkey request options, the response no longer tells accounts apart.

![What decoy responses can do, what they need and what they cannot do](https://s3.eu-central-1.amazonaws.com/corbado-cloud-staging-website-assets/passkey_login_enumeration_option_b_decoys_b5d0601c0a.png)

Decoys only work if they are indistinguishable from real responses:

- **Stable:** The same username always gets the same decoy. The spec suggests pseudo-random
  values derived deterministically from the username, for example an HMAC of the username
  with a server secret.
- **Realistic shape:** Count, credential ID length and transports must look like real
  passkeys from your actual provider mix.
- **Same timing:** Identical lookups and a fixed response-time floor.
- **Same error:** A decoy assertion fails exactly like a failed real one.
- **All endpoints:** Password, OTP, reset and registration must answer alike, now and after
  every release.

Decoys also have hard limits:

- **Real IDs stay exposed.** Real users still receive their real credential IDs before
  login, so the provider and correlation hints from section 2.2 remain for real accounts.
- **Everyone gets a dialog.** Users without a passkey see a passkey dialog they cannot
  complete, unless the passkey route starts behind a button.
- **Consistency is a permanent cost.** Every endpoint and every release has to keep the
  decoys intact, which adds a lasting testing burden.

Decoys fit teams that must keep a username-first page and can carry the engineering cost.
For most, the usernameless route is easier to defend.

## 7. Option C: Usernameless with One-Tap

For organizations with a strict enumeration rule, a usernameless login with a one-tap
passkey button gives up the least passkey usage. It combines three entry points on the
first login screen:

1. **Passkey autofill** in the username field via conditional UI.
2. **One-tap** for browsers with a cookie from an earlier passkey login.
3. **A generic passkey button** for users on a new device or with an expired cookie, placed
   next to the password login.

Design rules that keep it enumeration-safe:

- **No response depends on a typed username.** The password path behaves identically for
  every username and no passkey prompt follows a typed username.
- **Keep the cookie opaque.** Store a random handle that only your server can map to the
  account, with a defined expiry. It must never replace user verification. If the one-tap
  button shows a masked identifier, that is itself a disclosure on shared devices, which
  [Remember Me & Passkeys](https://www.corbado.com/blog/remember-me-passkeys) discusses in detail.
- **Plan for browsers without conditional UI.** Some browsers and in-app browsers do not
  support it. The button and the password path must still work without revealing anything.
- **Register discoverable credentials.** Passkeys created with `residentKey: required` work
  in both login models, so the model can change later. Switching from usernameless to
  username-first then changes how passkeys are used, while existing passkeys keep working.
  See [discoverable credentials](https://www.corbado.com/blog/webauthn-resident-key-discoverable-credentials-passkeys)
  for the details.

Users without a passkey on the device who press the generic button see a QR code or a "no
passkeys available" message from the operating system. Label the button clearly ("Sign in
with a passkey") and route a cancel or error back to the password field without a dead end.

Going usernameless also moves fraud holds, lock status and per-account risk checks from the
identifier step to after the verified assertion.
[Usernameless Passkey Login: Fraud & Lock Checks](https://www.corbado.com/blog/usernameless-passkey-login-fraud-checks)
shows where each control moves.

![Options A, B and C compared on what the user sees, enumeration, passkey usage and main cost](https://s3.eu-central-1.amazonaws.com/corbado-cloud-staging-website-assets/passkey_login_enumeration_options_compared_bf67931de6.png)

## 8. How the Trade-off shifts over the Rollout

The enumeration signal and the cost of going usernameless both depend on how many users
still have a phishable fallback. The following overview shows how both change from launch
to a passkey mandate.

Two consequences matter for the decision:

- **The disagreement is largest at launch.** Most users still have a password, the signal is
  most valuable to an attacker and the usernameless cost is highest because password habits
  and password autofill compete with the passkey.
- **Under a mandate, the question almost disappears.** Without a password to fall back to, a
  generic passkey button competes with nothing, and the signal points an attacker only to
  recovery.

## 9. Decision Checklist for Security and Product

Splitting the question into separate decisions helps both teams:

1. **Security position:** Can any validated combination of controls make a residual signal
   acceptable? Document the answer.
2. **Protection coverage:** What does the current edge and fraud tooling detect in
   enumeration tests?
3. **Usage gap:** How large is the gap for your audience, given platform mix and login
   frequency? Start with the ranges in section 4.
4. **Forecast:** Does the adoption forecast reflect the chosen login model, and has the
   business owner accepted it before launch?
5. **One-tap:** Is a one-tap passkey button planned, whichever model wins?
6. **Measurement:** Do you track button presses, autofill selections, one-tap impressions
   and cancellations from the client side? Server logs stop at the WebAuthn API boundary
   and cannot show which entry point users tried. See [passkey analytics](https://www.corbado.com/blog/passkey-analytics)
   for the metrics.
7. **Review date:** When will you revisit the model as enrollment grows or a mandate comes
   closer?

## 10. How Corbado can help

Corbado helps enterprises base this decision on measured data.
[Corbado Connect](https://docs.corbado.com/corbado-connect) adds passkeys on top of an
existing identity provider, including conditional UI, a one-tap passkey button and passkey
intelligence for start rules in either login model.
[Corbado Observe](https://www.corbado.com/blog/authentication-observability) instruments the login page from the
client side, so teams can see how often users press the passkey button, select autofill or
use one-tap, and how the passkey login rate develops by operating system and browser after
a change of login model.

[Video: Login methods](https://www.corbado.com/videos/features/login-methods.mp4) ([See Method Uplift in Corbado Observe](https://www.corbado.com/observe/method-uplift))

## 11. Conclusion

Account enumeration in passkey login is a trade-off you can quantify. A username-first
flow reveals through the `allowCredentials` list which accounts exist and have a passkey.
Mitigations make that signal more expensive to harvest and leave it in place. Decoys can
hide it at a permanent consistency cost. A usernameless login avoids it, at a price in
passkey usage that depends mostly on whether a returning-browser shortcut such as a one-tap
passkey button is available.

For organizations with a strict enumeration rule, launching usernameless with one-tap is
usually the most defensible starting point. Recalculate the forecast, measure from the
client side and review the decision as the phishable fallbacks disappear.

## Frequently Asked Questions

### Does a username-first passkey login allow account enumeration?

Yes, if the server only returns WebAuthn request options with a populated
`allowCredentials` list for accounts that have a passkey. An outsider can submit any
username and read whether the response contains credential IDs. That reveals that the
account exists and has a passkey, and the number and length of the IDs can hint at the type
of authenticator.

### Is conditional UI (passkey autofill) safe against account enumeration?

From the server's perspective, yes. Conditional UI starts a WebAuthn request with an empty
`allowCredentials` list, so the server response is identical for every visitor. The browser
shows matching passkeys locally and the server learns nothing until the user selects a
passkey and completes user verification.

### How much passkey usage does a usernameless login cost?

It depends mostly on whether a returning-browser shortcut such as a one-tap passkey button
is available. With one-tap, usernameless login typically trails a username-first login with
one-tap by around 10-20% of passkey logins. Without it, usernameless relies on passkey
autofill and a generic button that only around 10-25% of passkey holders press, and the gap
can reach half of the passkey logins.

### Can decoy credential IDs prevent passkey account enumeration?

W3C WebAuthn Level 3 suggests returning plausible imaginary values for unknown usernames,
derived deterministically from the username. This hides account existence if the decoys
match real responses in count, ID length, transports, timing and error behavior on every
endpoint. Decoys do not hide the real credential IDs of real users and they show a passkey
dialog to users without a passkey.
