Reference Guide

Passkey Management

Five stores, three of which sync, none of which fully interoperate — with recovery as the unsolved middle. Where a passkey lands is decided the first time you create one, and migration is the part that still does not work.

← Back to Reference Hub

Best for: Someone who is genuinely all-Apple and intends to stay that way.

  • Sync scope: Apple devices only, end-to-end encrypted
  • Requires: 2FA on the Apple Account plus a device passcode
  • Managed in: the standalone Passwords app (iOS/iPadOS/macOS 18+)
  • Third-party AutoFill providers supported, several at once
  • CXP import and export shipped in iOS/macOS 26 — biometric-gated, app-to-app
  • Attestation: none — consumer passkeys report a zeroed AAGUID

Limitations: On Windows it is reachable only through iCloud Passwords plus a browser extension — there is no OS integration. The zeroed AAGUID means an enterprise relying party cannot verify which authenticator holds the key, so iCloud Keychain is structurally unusable where attestation is a requirement.

SyncsApple onlyCXP on iOS 26

Best for: Windows users who understand that "Microsoft passkeys" is not one thing.

  • (1) Windows Hello — device-bound, Hello container, no sync
  • (2) Microsoft Password Manager — syncs across Windows via a personal Microsoft account
  • (3) Third-party as system provider — KB5068861, Windows 11 24H2 and 25H2
  • In mode 3, Hello does the biometric unlock; the third-party app holds the credential
  • Enterprise: Entra passkeys in Windows Hello reached GA from late April 2026 (worldwide ~mid-June; GCC High/DoD Oct 2026)
  • Microsoft mandates passkeys in Entra by February 2027 and is retiring SMS and voice MFA

Limitations: Entra passkeys in Windows Hello are device-bound only — register per device, no sync, no attestation support, and they do not cover device sign-in. Windows Hello for Business remains the recommendation for managed devices. Microsoft has shipped no export mechanism on any platform, and third-party provider mode needs the full desktop app rather than a browser extension.

Three storesNo exportEntra deadline Feb 2027

Best for: Anyone whose common denominator across machines is Chrome rather than an operating system.

  • Runs on all six platforms: Android, ChromeOS, Windows, macOS, iOS/iPadOS, Linux
  • Rides Chrome, not the OS — the browser is the common denominator
  • Android 14+ can hand the default to another provider via Credential Manager
  • Android 17 shipped CXP: ProviderEventsManager importCredentials / registerExport
  • Theft hardening: "Mark as lost" in Find Hub, harder PIN throttling
  • Advanced Protection uses FIDO2 credentials as the enrollment baseline

Limitations: On iOS it only works if Chrome is set as an autofill provider in System Settings — otherwise passkeys land in iCloud Keychain and the user usually does not notice. On macOS the GPM-or-Keychain choice is made per credential, which is how a collection ends up split. See the callout below for the August 2026 Unit 42 research into GPM sync-key handling on Windows.

SyncsAll six platformsSee security note

Best for: Anyone who lives on more than one ecosystem — this is the configuration that survives contact with reality.

  • Indifferent to platform — the same vault on Windows, macOS, Linux, iOS, and Android
  • System-provider status: Windows 11 (Nov 2025), Android 14+, iOS AutoFill
  • 1Password and Bitwarden co-authored the FIDO Alliance CXP specification
  • Smaller managers — Bitwarden, Proton Pass, KeePass — still allow plaintext file export
  • Recovery runs through the vault password and the recovery kit

Limitations: The trade-off is not that the single point of failure goes away, it is that it moves — from the platform vendor to the manager vendor. The vault password and recovery kit become the master key for everything. The majors deliberately do not offer plaintext export, so leaving a major manager on desktop is currently not a supported operation.

SyncsCross-platformVendor is the master key

Best for: Break-glass recovery, and any enterprise that needs to prove which authenticator holds the key.

  • Device-bound: no sync, no cloud, nothing to breach remotely
  • Attestation that an enterprise relying party can verify
  • Storage is finite: ~25 discoverable credential slots on older YubiKey 5s, 100 on firmware 5.7+
  • A human present to touch the key is a workable path for supervised automation
  • Standing advice: enroll two keys everywhere

Limitations: No recovery. Lose the key and the credentials on it are gone — which is why the two-key rule is not optional advice. When the slots fill up, adding a credential means deleting one.

Device-boundAttestationNo recovery

Best for: Galaxy owners who want to know why they were asked three times where to save a passkey.

  • Stores passkeys and plugs into the Android Credential Manager
  • Now under the Samsung Wallet umbrella
  • On a Galaxy it competes with GPM and any installed third-party manager
  • Behaves as a standard Credential Manager provider to the relying party

Limitations: It is a fourth default on a device that already had three. The risk is not the store itself but the split collection that results when the user picks a different provider each time the prompt appears.

Android onlyFourth default

Best for: Workforce credentials, where the requirements are the inverse of the consumer ones.

  • Requirement is inverted: device-bound plus attestation, not synced plus convenient
  • Attestation proves which authenticator holds the credential
  • Entra is both an IdP here and one of the three Microsoft stores
  • Hardware keys are the store that satisfies the enterprise requirement cleanly

Limitations: The result is two worlds that cannot see each other: workforce credentials and the same employees’ personal passkeys live in separate systems with no bridge. An enterprise cannot attest a consumer store, and a consumer store will not become attestable, because the zeroing is the feature.

Device-boundAttestationCannot see personal passkeys

The August 2026 Unit 42 research into Google Password Manager on Windows

In August 2026, Palo Alto Unit 42 published three attack paths against Chrome’s Google Password Manager on Windows machines with a TPM. Pass-ta-key extracts Chrome’s device identity key and signs requests without the User Verified flag. Silver Pass-ta-key forces re-enrollment in order to install an attacker verification key. Golden Pass-ta-key extracts the 32-byte Security Domain Secret from Chrome memory and decrypts synced passkeys. State the scope precisely in both directions: all three require malware already running on the machine, so this is a post-compromise escalation rather than a remote attack — and equally, no CVEs have been assigned, Chrome 142 was vulnerable, only partial fixes are documented, and Google has not publicly commented. This does not break passkeys as a standard. It breaks Google Password Manager’s sync-key handling on Windows, and that distinction is the whole point.