---
url: 'https://www.corbado.com/blog/are-passkeys-quantum-safe'
title: 'Are Passkeys quantum safe? ML-DSA Passkeys explained'
description: 'Are passkeys quantum safe? What a quantum computer breaks in a passkey, where ML-DSA passkeys stand in 2026 and how to test one in your browser.'
lang: 'en'
author: 'Vincent Delitz'
date: '2026-11-19T18:27:08.989Z'
lastModified: '2026-11-19T19:46:51.726Z'
keywords: 'are passkeys quantum safe, post-quantum passkeys, ml-dsa passkeys, quantum resistant passkeys, webauthn ml-dsa, pqc passkeys'
category: 'WebAuthn Know-How'
---

# Are Passkeys quantum safe? ML-DSA Passkeys explained

## Key Facts

- **Today's passkeys are not quantum safe:** almost all of them sign with ES256 (ECDSA on
  P-256), which a large enough quantum computer could break.
- **The risk is narrower than for certificates:** a passkey's public key sits in the
  website's database, not in public. An attacker first needs a leak of that database.
- **Post-quantum passkeys exist:** ML-DSA has COSE identifiers since RFC 9964 (May 2026)
  and 1Password creates ML-DSA passkeys since version 8.12.44 (October 2026).
- **Support is thin:** Apple Passwords in Safari 27 still returns an ES256 passkey when a
  website asks for ML-DSA only. The FIDO Metadata Service lists only one
  uncertified authenticator with ML-DSA.
- **You can test it now:** use the
  [post-quantum preset of the Passkeys Debugger](https://www.passkeys-debugger.io/?preset=create-post-quantum-passkey).

## 1. Introduction

Quantum security advisors increasingly put passkeys on the list of things that become
"forgeable" once a cryptographically relevant quantum computer exists, next to digital
identity wallets, single sign-on tokens and e-signatures. Their point for identity teams:
almost every digital identity today rests on a digital signature and the algorithms behind
most of them will not survive "Q-Day".

That is correct for the algorithm. The details matter for what you should do about it,
though, because a passkey is used differently from a certificate or a signed document.
This article answers three questions:

1. What exactly would a quantum computer break in a passkey?
2. How far are post-quantum passkeys in November 2026?
3. What should a relying party do now and how can you test it yourself?

## 2. What a Quantum Computer would break in a Passkey

### 2.1 ES256 and Shor's Algorithm

A [passkey](https://www.corbado.com/glossary/passkey) is a key pair. The private key stays with the authenticator
(a password manager, the platform or a hardware key). The public key goes to the
[relying party](https://www.corbado.com/glossary/relying-party) at registration. At login the authenticator signs
a fresh challenge and the server checks the signature with the stored public key.

Which algorithm is used is decided at registration through `pubKeyCredParams`. In practice
almost every passkey today uses **ES256** (ECDSA on the P-256 curve), with RS256 and EdDSA
as the other common options. Our article on
[pubKeyCredParams and credentialPublicKey](https://www.corbado.com/blog/webauthn-pubkeycredparams-credentialpublickey)
covers how that negotiation works.

ECDSA, EdDSA and RSA all rely on problems that Shor's algorithm solves efficiently on a
large enough quantum computer. With such a machine, an attacker who knows a public key can
compute the matching private key and then produce valid signatures.

### 2.2 Where the Public Key lives

This is where passkeys differ from the examples usually given for "trust now, forge later"
(collect public keys today, forge signatures once quantum computers exist).
A certificate, a firmware signing key or an e-signature is verified with a public key that
is published on purpose. Anyone can collect it.

A passkey's public key is only stored by the website it was created for. To forge a login,
an attacker therefore needs two things:

- the public key, which means a breach or leak of the relying party's credential table
- a quantum computer that can derive the private key from it

That makes the realistic threat **"steal public keys now, forge later"**. A credential table
leaked in 2026 could be used to log in as those users years later if the passkeys are still
registered by then. Public keys are usually treated as harmless and stored with weaker
protection than password hashes. Once quantum attacks are practical, that assumption no
longer holds.

### 2.3 What does not apply to Passkeys

**Harvest now, decrypt later:** this is a threat to encrypted data, not to login
signatures. A recorded passkey assertion is useless to an attacker: it signs a one-time
challenge and contains no secret. The transport around the login is a separate question and
browsers already protect it:
[Chrome has used hybrid ML-KEM key exchange in TLS by default](https://security.googleblog.com/2024/09/a-new-path-for-kyber-on-web.html)
since version 131 in November 2024.

**Phishing:** this does not come back either. A forged signature still has to be produced
for the right relying party ID, so the binding of a passkey to its domain stays intact.

### 2.4 Hardware Keys and Attestation

Two parts of the passkey ecosystem live longer than a typical credential:

- **Hardware security keys:** they cannot usually update their firmware. A key bought today
  with ES256 only will never sign with ML-DSA. Organisations that hand out keys for five or
  more years should factor that into model choice.
- **Attestation roots:** vendors use long-lived public keys to sign their authenticator
  certificates. If one of them is broken, an attacker could make a fake authenticator look
  certified. Relying parties that enforce
  [attestation or AAGUID allowlists](https://www.corbado.com/blog/restrict-passkey-authenticators-aaguid) depend
  on those roots.

## 3. When does this become urgent?

Nobody knows when a cryptographically relevant quantum computer will exist. The
[Global Risk Institute's Quantum Threat Timeline Report 2025](https://globalriskinstitute.org/publication/quantum-threat-timeline-report-2025b/)
summarises expert estimates: such a machine is "quite possible" (28 to 49%) within the next
ten years and "likely" (51 to 70%) within 15. Resource estimates keep falling as well:
1Password's [launch post for post-quantum passkeys](https://1password.com/blog/post-quantum-passkeys)
cites a 2026 estimate of about 26,000 physical qubits for breaking elliptic curve
signatures.

Regulators do not wait for a date. The reference dates most identity teams plan
against are already set:

| Source | Deadline |
|---|---|
| [NIST IR 8547](https://csrc.nist.gov/pubs/ir/8547/final) (US, final October 2026) | All quantum-vulnerable signature algorithms, including ECDSA P-256, deprecated after 2030 and disallowed after 2035 |
| [EU Coordinated Implementation Roadmap](https://digital-strategy.ec.europa.eu/en/library/coordinated-implementation-roadmap-transition-post-quantum-cryptography) (June 2025) | Start the transition by the end of 2026, high-risk use cases no later than 2030, as many systems as feasible by 2035 |
| [Microsoft](https://www.microsoft.com/en-us/security/blog/2026/06/30/microsoft-advances-quantum-safe-security-as-the-risk-timeline-shifts/) (June 2026) | Transition products and services to post-quantum cryptography by 2029 |

For passkeys, the practical consequence is simple: credentials registered from now on may
well still be in use when the first deadlines hit. Starting with post-quantum options for
new registrations costs little and shortens the later migration.

## 4. State of post-quantum Passkeys in November 2026

The post-quantum signature algorithm for passkeys is **ML-DSA**
([FIPS 204](https://csrc.nist.gov/pubs/fips/204/final), formerly Dilithium). It comes in
three parameter sets, registered for COSE in
[RFC 9964](https://www.rfc-editor.org/rfc/rfc9964.html#name-new-cose-algorithms):

| Variant | COSE ID | Public key | Signature |
|---|---|---|---|
| ES256 (today) | -7 | 64 bytes | about 70 bytes |
| ML-DSA-44 | -48 | 1,312 bytes | 2,420 bytes |
| ML-DSA-65 | -49 | 1,952 bytes | 3,309 bytes |
| ML-DSA-87 | -50 | 2,592 bytes | 4,627 bytes |

```mermaid
---
config:
  xyChart:
    width: 900
    height: 450
---
xychart-beta
    title "Signature size per login in bytes"
    x-axis ["ES256", "ML-DSA-44", "ML-DSA-65", "ML-DSA-87"]
    y-axis "Bytes" 0 --> 5000
    bar [70, 2420, 3309, 4627]
```

The sizes matter: credential tables, session payloads and logging need room for keys and
signatures that are 20 to 65 times larger than today.

Where the ecosystem stands:

| Layer | Status |
|---|---|
| Standard | [RFC 9964](https://www.rfc-editor.org/rfc/rfc9964.html) (May 2026) defines ML-DSA for COSE with a new key type (AKP). WebAuthn can carry it without a spec change. |
| Server libraries | [SimpleWebAuthn 14](https://github.com/MasterKale/SimpleWebAuthn/releases/tag/v14.0.0) verifies ML-DSA-44/65/87. In runtimes that support it, its registration options put ML-DSA-44 first, so upgrading the library can change what users get. |
| Password managers | [1Password 8.12.44](https://1password.com/blog/post-quantum-passkeys) (October 2026) creates ML-DSA passkeys when a website asks for them. Not yet on iOS or through the native macOS autofill extension. Existing passkeys are not converted. |
| Apple Passwords | In our tests with Safari 27, a request for ML-DSA only still returns an ES256 passkey. WebAuthn requires the relying party to reject that registration. |
| Chrome | In our tests, the DevTools virtual authenticator in Chrome 157 supports ML-DSA-44/65/87, which makes testing possible without hardware. |
| Hardware keys | The FIDO Metadata Service lists one authenticator with ML-DSA-65 in its capabilities (CanoKey Canary, not certified). The FIDO registry has no ML-DSA name for authentication algorithms yet. |

In short: the standards are ready, one major password manager ships it and the rest of the
ecosystem has not caught up.

## 5. What Relying Parties should do now

**1. Know what you accept and what you store:** check which algorithms your registration
requests and which ones your credential table actually contains. Most teams find ES256 and
RS256 only.

**2. Add ML-DSA, keep the classic algorithms:** authenticators take the first algorithm in
`pubKeyCredParams` that they support. Listing ML-DSA first and ES256 after it means that
ML-DSA-capable authenticators create post-quantum passkeys and everyone else still gets a
working ES256 passkey. Dropping ES256 would lock out almost all users today. Before you do
this, make sure your server verifies ML-DSA and your storage can hold the larger keys.

**3. Plan for re-registration:** there is no way yet to upgrade an existing passkey to a
new algorithm in place. 1Password says it is engaged with the W3C and the FIDO Alliance on
one. Until then, moving a user to a post-quantum passkey means creating a new credential.
[Automatic passkey upgrades](https://www.passkeys-debugger.io/how-to/automatic-passkey-upgrade-test)
after a successful login and the [Signal API](https://www.corbado.com/blog/webauthn-signal-api) for cleaning up
old credentials are the tools for doing that quietly.

**4. Protect the public key table:** treat a leak of passkey public keys as a future
account takeover risk and store them with the same care as other credential material.

**5. Ask your vendors:** ask your identity provider and your hardware key vendor for their
ML-DSA roadmap. If your provider has no date, that is your answer.

## 6. Test a post-quantum Passkey yourself

The [Passkeys Debugger](https://www.passkeys-debugger.io) has a preset that asks for an
ML-DSA passkey and nothing else:

**[Create a post-quantum passkey (ML-DSA)](https://www.passkeys-debugger.io/?preset=create-post-quantum-passkey)**

The preset requests ML-DSA-87, ML-DSA-65 and ML-DSA-44 in that order and deliberately has
no ES256 fallback, so you see exactly what your authenticator does:

```mermaid
flowchart TD
    classDef ok fill:#e8f5e9,stroke:#2e7d32,stroke-width:2px;
    classDef neutral fill:#e3f2fd,stroke:#1565c0,stroke-width:2px;
    classDef bad fill:#ffebee,stroke:#c62828,stroke-width:2px;
    Start["Debugger preset requests<br/>ML-DSA-87, ML-DSA-65, ML-DSA-44<br/>no ES256 fallback"]:::neutral
    Start --> Q{"What does the<br/>authenticator do?"}
    Q -->|"Supports ML-DSA"| A["Post-quantum passkey created<br/>strongest variant, atom icon"]:::ok
    Q -->|"No ML-DSA support"| B["Ceremony fails<br/>correct behaviour"]:::neutral
    Q -->|"Creates ES256 anyway"| C["Algorithm tile turns red<br/>server must reject it"]:::bad
```

- **Your authenticator supports ML-DSA:** the result shows the strongest variant it offers,
  marked with an atom icon, along with the decoded public key and its size.
- **It does not:** the ceremony fails, which is the correct behaviour.
- **It creates a classic passkey anyway:** the algorithm tile turns red, because the key
  was not in the requested list and a server has to reject it. This is what Apple Passwords
  in Safari 27 does today.

To test without a supporting password manager, open Chrome DevTools, go to the WebAuthn
panel and add a virtual authenticator. In our tests with Chrome 157 it supported all three
ML-DSA variants. With 1Password 8.12.44 or later on desktop, the same preset creates a real
post-quantum passkey.

## 7. How Corbado can help

A post-quantum migration starts with knowing which authenticators your users hold, because
that decides who can get an ML-DSA passkey today and who needs a fallback.
[Corbado Observe](https://www.corbado.com/observe), the
[authentication observability layer](https://www.corbado.com/blog/authentication-observability), shows this for
your existing login without changing your identity provider.

[Video: Authenticator inventory](https://www.corbado.com/videos/features/authenticator-inventory.mp4) ([See Authenticator Inventory in Corbado Observe](https://www.corbado.com/observe/authenticator-inventory))

- **Authenticator Inventory:** see which authenticator models and providers your users
  have registered, synced and device-bound, so you can size how many users 1Password,
  Apple Passwords or hardware keys represent before you change `pubKeyCredParams`.
- **Passkey Errors:** when you add ML-DSA to registration, compare errors by browser and
  operating system to catch authenticators that reject or mishandle the new request.
- **Login funnel:** follow a re-registration campaign through each step and see where users
  stop.

## 8. Conclusion

Passkeys are not quantum safe today and the claim that they become forgeable after Q-Day
is correct for the algorithm. The attack is harder than for certificates, because the
attacker first needs the relying party's public keys. It also does not reopen phishing.

The fix already exists. ML-DSA is standardised for passkeys, one major password manager
creates them and server libraries verify them. What is missing is broad authenticator
support. Relying parties can prepare now by accepting ML-DSA alongside ES256, making room
for larger keys and protecting their credential tables. New passkeys then become
post-quantum as soon as users' authenticators can create them.

## Frequently Asked Questions

### Can a quantum computer use a recorded passkey login to break into an account?

No. A recorded passkey assertion only signs a one-time challenge and contains no secret, so harvest now, decrypt later does not apply to login signatures. The realistic quantum threat is steal public keys now, forge later: an attacker needs a leaked credential table plus a quantum computer that derives the private keys from those public keys.

### Which COSE algorithm identifiers do ML-DSA passkeys use in WebAuthn?

RFC 9964 registers ML-DSA-44 as -48, ML-DSA-65 as -49 and ML-DSA-87 as -50, together with the new AKP key type. WebAuthn carries these identifiers in pubKeyCredParams without a spec change. An ML-DSA-44 signature is 2,420 bytes compared with about 70 bytes for ES256.

### How should a relying party order pubKeyCredParams to support post-quantum passkeys?

List ML-DSA first and keep ES256 after it. Authenticators pick the first algorithm they support, so ML-DSA-capable authenticators such as 1Password 8.12.44 create post-quantum passkeys while all others still create working ES256 passkeys. Dropping ES256 today would lock out almost all users.

### What deadlines apply to migrating ECDSA P-256 signatures to post-quantum algorithms?

NIST IR 8547 deprecates quantum-vulnerable signatures such as ECDSA P-256 after 2030 and disallows them after 2035. The EU roadmap expects high-risk use cases to move no later than 2030. Microsoft targets 2029 for moving its products and services to post-quantum cryptography.

### Can an existing passkey be upgraded to ML-DSA?

No. There is no in-place algorithm upgrade for an existing passkey yet, so moving a user to ML-DSA means registering a new credential, for example through an automatic passkey upgrade after a successful login. The WebAuthn Signal API then tells the passkey provider to remove the old one. 1Password is working with the W3C and the FIDO Alliance on an upgrade path.
