#80 - Understanding Passkeys: Benefits And Limitations
A passkey blocks a fake login page, but an enabled SMS fallback leaves another route into the account.
17 min read
Passkeys are gaining attention as a new way to log in without passwords. How do they work, and how do they compare to traditional multi-factor authentication (MFA)? In this episode, we explore the history of passwords, the strengths and weaknesses of common MFA methods, and the potential of passkeys to enhance security. What threats do passkeys mitigate, and what still remain?
Adding a passkey doesn’t close your weaker login paths
Sooner or later a site will offer you a passkey. The dialog asks whether to save it, and for most people the honest answer is that they don’t know where it’s going, what it replaces, or whether anything just got better. Click yes and the login does get faster, and a convincing copy of that login page stops working on you. What the click doesn’t do is close the older ways into the same account, and those usually stay switched on.
That matters because phishing is how accounts get taken now. People we work with get credible messages every week: the accountants, the people with cloud access, the ones whose names are on the website. Verizon’s 2025 Data Breach Investigations Report puts credential abuse at the top of the initial access vectors at 22%, ahead of vulnerability exploitation at 20% and phishing at 16%, and it notes that credential abuse and phishing overlap, because a credential that works may well have come from a phishing attempt nobody recorded. A passkey is a factor that a convincing copy of a login page can’t harvest. It’s also a factor whose benefit you can cancel out by leaving a text message switched on.
Passwords have had a long run. The password is usually credited to MIT’s Compatible Time-Sharing System around 1961, where Fernando Corbató’s team gave each researcher an account so they couldn’t read each other’s files (MIT News). More than sixty years later, the shape of the problem is unchanged: a password is a secret made of knowledge, and knowledge can be handed over. Everything since has been an attempt to bolt something harder onto it.
Codes you type can be typed into the wrong box
One secret was never going to hold, so the industry bolted on a second factor, and most of the common ones still end in a number you read off a screen.
Which second factors can a phishing page defeat?
All of the ones that end with you typing that number somewhere, which covers SMS, authenticator apps and emailed codes alike.
SMS one-time codes rest on something you have, your SIM card, and they work on any phone, which is why they’re still everywhere. Two things break them. The first is the SIM swap: someone calls your carrier’s support line, convinces them they’re you, and asks for the number to be moved to a new SIM. There’s an industry behind it, with one person on the phone to the carrier and another holding the handset that receives the code. The FBI logged 1,611 SIM swap complaints in 2021 with adjusted losses over $68 million, against 320 complaints and roughly $12 million across the whole of 2018 to 2020 (IC3 alert I-020822-PSA). Second is the network itself: SS7, the signalling stack behind mobile text, was not built with this threat in mind.
NIST noticed a while ago. Since the 2017 revision of SP 800-63B, use of the phone network for out-of-band verification is marked RESTRICTED, and a provider that keeps it has to offer at least one alternative that isn’t restricted, warn subscribers about the risk, and write a migration plan (NIST SP 800-63B). That’s not a ban. It’s a standards body saying the clock is running.
Generated codes are better. A TOTP code is computed on your device from a secret you share with the site, so no message crosses the mobile network for anyone to intercept (RFC 6238). Protection moves rather than disappears: the seed still has to reach your authenticator app, usually through a QR code, and the code still travels when you submit it. Neither of those is the hard part for an attacker. A page that looks like your bank asks for the password, then asks for the six digits, and relays both while they’re still valid.
Emailed codes have the same property plus a dependency. The magic link version, where a site mails you a link instead of a code and skips the password entirely, is a decent system as far as it goes. It also means your inbox is the last line of defense, which comes back later on this page.
Approvals and hardware fobs move the weak point
Stop typing codes and the obvious next step is to tap a button instead.
Is approving a push notification good enough?
It’s a real improvement over a code you can be talked into reading aloud, and it has a known failure that needs no technical skill at all.
Push approval, through Duo or Microsoft Authenticator or Google’s prompt, sends a notification to a device you’re already signed in on. The attacker who already has your password sends login attempt after login attempt until the notifications stop being a security decision and start being an annoyance. It’s called MFA fatigue, or push bombing, and it works often enough to be standard practice. CISA’s fact sheet lists push bombing alongside phishing, SS7 and SIM swap as the reasons to move off these methods, and recommends number matching as an interim fix where phishing-resistant MFA isn’t available yet (Implementing Phishing-Resistant MFA, October 2022).
Hardware code generators sit at the other end. An RSA SecurID fob is tamper resistant, tied to one account, and has no network to attack. What the hardware protects is the generator, not the exchange: the digits still leave through your keyboard or your mouth, so the relay attack above works on them unchanged. Its other cost is friction. No fob, no login, and we have heard stories of teams keeping them in a safe, so logging in means someone walks over, opens it, and reads the digits down the phone to whoever needs them.
The pattern across all of these is the same. Each one asks a tired human to make a judgment call at the exact moment an attacker has engineered pressure. The factor that finally breaks the pattern is the one that takes the judgment away.
What a passkey actually is
That factor is the passkey, and it pays to be precise about it, because the marketing is vague in both directions.
What does a passkey do that a one-time code can’t?
It signs a challenge with a private key that is never disclosed to the site, and it refuses to sign anything for the wrong domain.
The mechanics are the ones a FIDO security key has used for years. When you register, your authenticator generates a key pair for that site specifically. The public key goes to the site, and the private key stays with the authenticator: the site never sees it, so there is nothing in its database to steal and nothing for you to retype. Where that private key lives is the part worth asking about. A device-bound passkey stays in one protected store, the Secure Enclave on a Mac, the equivalent chip on a phone, or a hardware key you plug in. A synced passkey is copied between your devices in encrypted form. NIST’s April 2024 supplement allows that when key generation, storage and access control meet its requirements (SP 800-63B supplement 1). Either way, at login the site sends a challenge, your authenticator verifies you locally with a fingerprint, a face or a PIN, then signs. WebAuthn keeps that step apart from user presence: a touch on a security key shows only that somebody is at the device, while user verification is the check that the somebody is you.
The part that stops phishing is the binding. A WebAuthn credential is scoped to a relying party id, effectively the site’s domain, and the browser checks the origin it’s talking to before it will ask your authenticator for anything (W3C WebAuthn Level 2).
There’s a second benefit, on the site’s side of the connection. A site has to hold something derived from your password, hashed and salted if they’ve done it right, and that store is what gets dumped in a breach. With passkeys the site holds a public key. It was never a secret, and a stolen copy of the user table doesn’t let anyone sign in.
None of this is new cryptography. It’s public-key signatures, the same mechanism behind TLS certificates and SSH, where you hand GitHub a public key and keep the private one. What changed is the packaging: a user-friendly front end on a well-tested idea, which beats a clever new one.
An account is only as strong as its weakest enabled login
That’s the theory. The practice depends entirely on what else is still switched on.
Does adding a passkey make an account harder to break into?
It cuts your phishing exposure from the first time you use it, and every weaker method still enabled stays open to an attacker who can meet that method’s requirements.
Most sites do this halfway. You sign in with what you already have, then register a passkey as another option to sign in. Another option, not a replacement. The old password still works. The SMS fallback still works. The recovery flow still works.
Both halves of that are real. Phishing protection applies the moment you start using the passkey, whatever else is registered, and the login is faster with no code to copy. The SIM swap path is also still open, to an attacker who has the password to go with it.
Two practical consequences. First, go through the security settings of the accounts you care about and delete SMS wherever a stronger method exists, rather than stacking the new one on top. Second, before deleting anything, register a second passkey or a second hardware key, because the fallback you’re removing was also your way back in.
One thing a passkey doesn’t fix: the session that comes after. Authentication ends when the cookie is issued, and a stolen session cookie is as good as a login. That problem is untouched by any of this, and it stays yours.
Where does the passkey go when you click save?
Which of those paths you can close depends on where the credential ended up, decided by a dialog nobody reads.
Where does a passkey actually get stored?
In whichever of four places wins that dialog: the browser, the operating system’s keychain, your password manager, or a hardware key you plug in.
The dialog is genuinely bad: it often won’t say which store is about to take the credential, and the browser volunteers hardest. Our objection to the browser is really about passwords. Whoever gets into your Google account can reach the passwords saved there, unless Chrome’s optional sync passphrase is set: with one, they aren’t readable at passwords.google.com either (Google Chrome Help). Passkeys there work differently. Google Password Manager keeps them end to end encrypted, and a new device needs the Password Manager PIN or your Android screen lock before they sync down (Google, September 2024).
The operating system is the common answer now, and it comes with a compromise. Early FIDO credentials were non-transferable by design: one device, one key. Apple and Google both added syncing within their own ecosystem, because non-recoverable credentials are a support disaster. Apple’s version encrypts keychain items end to end, so they cross Apple’s servers unreadable (iCloud Keychain security overview). It’s still a trade: anything transferable can be taken.
Password managers store passkeys too, and the counting goes wrong in both directions. A synced passkey in a vault is still a key you possess plus a local check that you’re you. NIST’s supplement accepts that arrangement at AAL2 when the sync fabric is configured properly. What you lose is independence. Put the password, the TOTP seed and the passkey for an account in one vault behind one master password, and one compromise produces all three.
Hardware keys remain the strongest option and the least forgiving. The secret can’t be read back off the device, which is the point, and why there’s no backup to restore. Yubico’s own answer is to register a second key at the same time as the first (Can I duplicate a YubiKey?). Capacity used to be the catch: older keys held 25 discoverable credentials, and firmware 5.7 raised that to 100 (Yubico, May 2024).
Mattias’s advice is to order two keys today and start with the sites that support them. Paulina won’t: she lost five at a previous employer, and a factor you leave in a drawer is a factor you can’t use.
Recovery is the way in, and the way out for your family
Whichever store you pick, there is a path into the account that never touches it.
What’s the real attack path into a passkey-protected account?
Often the recovery flow, so find out what yours actually does before you trust the passkey sitting in front of it.
For most consumer accounts the reset link lands in your inbox, which makes that inbox the last line of defense for everything else you own. Magic links have the same shape. The convenience and the attack path are one mechanism, and whoever reads your mail gets both.
What a reset gets an attacker varies by service, and the absolute cuts both ways. GitHub does not let a password reset stand in for the second factor. Lose your 2FA credentials there and you’re sent to recovery codes, a signed-in device, or an SSH key you can prove you hold (GitHub docs). An account with no recovery credential may not be recoverable at all. Other services are more relaxed. The design is per service, so read the recovery page for the accounts that could reset the rest, and treat what you find as the real strength of that account.
There’s a second kind of recovery nobody plans for, which is the one where you’re not available. Can the people close to you reach your bank, your mail and your password manager if something happens to you? Before one of us went under general anaesthetic for a procedure, a spouse got three things the day before: the phone PIN, the PIN for BankID, and the password manager’s master password. With those three, the rest was reachable. Do that exercise with the people who’d need it, and label the hardware while you’re at it. A security key or a crypto wallet in a drawer looks like a spare USB stick to anyone who didn’t buy it, and gets handed to the first person who asks.
Microsoft’s passwordless default, and what it settles
The reason this came up when it did is a change at consumer scale. As of 1 May 2025, brand new Microsoft accounts are passwordless by default, never enrolling a password at all, and existing users can delete theirs from account settings. Microsoft says the reworked sign-in experience cut password use by over 20% in its experiments.
An account with no password is an account that credential stuffing can’t touch, which is a real reduction in attack surface at that scale. It doesn’t settle the rest of this page. Deleting the password only helps if what remains is strong, and the recovery path is still the recovery path.
What this means for teams
If you take one thing from this, make it the inventory rather than the technology. Adding a factor is easy and feels productive. Removing one is the part that changes the outcome, and it’s the part that gets skipped.
This week:
- List every enabled sign-in and recovery method on your crown jewel accounts: the identity provider, the email domain, the cloud root accounts, the password manager, the source control org. Not the ones you use. All the ones that still work.
- Delete SMS as a factor wherever a stronger method is available, after registering a second passkey or a second hardware key so you still have a way back in.
- In the risk register, record a password and its passkey in one vault as a single point of compromise, not two independent factors.
- Read the recovery page for the accounts that could reset the rest, your email domain first, before you need it. Whether a password reset can stand in for the second factor is decided service by service, and GitHub’s answer is not every service’s.
Common questions, answered
Should I delete SMS two-factor after I add a passkey?
Delete it, but register a second passkey or a second hardware key first so you keep a way back into the account. SMS stays attackable through SIM swap and the phone network itself, and while it’s enabled an attacker with your password can use it instead of the passkey you just added. NIST has marked phone-network delivery as a restricted method since 2017. If a service won’t let you remove SMS, treat that account as protected at the level of SMS, whatever else is registered.
What happens if I lose the device with my passkeys on it?
It depends on which store holds them, so decide that before it happens. Passkeys in Apple’s or Google’s ecosystem sync to your other devices there, so a lost phone is an inconvenience, provided you can pass the provider’s check: Google asks for the Password Manager PIN or your Android screen lock on a new device. A passkey on a hardware key doesn’t sync and can’t be copied off, which is why Yubico’s advice is to register two keys at once and keep the spare elsewhere. After that you’re into the account’s recovery flow, worth reading before you need it.
Is it safe to keep passkeys in my password manager?
It’s a reasonable choice for most accounts, as long as you’re honest about what it buys. The passkey is still a key you hold plus a local check that you’re you, accepted at AAL2 when the vault’s sync is configured properly. What it isn’t is independent: password, code generator and passkey behind one master password means one unlock produces all three, and that makes the vault worth real effort to attack. Your email account deserves the hardest thinking either way, because your inbox can unlock accounts whose recovery process relies on email alone.
Related episodes
- #24 Ways To Protect Yourself From Data Breaches And Mitigate Consequences covers the account triage behind this one: which accounts are the crown jewels, and why you register a backup key before you need it.
- #15 Remote Work Security is where we first got into WebAuthn and what a YubiKey does and doesn’t move onto hardware.
- #73 Incident Response: Key Preparations You Need is the preparation episode we point to for the family version of this, including the code word for verifying someone on a call.
Resources
- Microsoft Security Blog, Pushing passkeys forward, 1 May 2025. Passwordless by default for new accounts, and password deletion for existing ones.
- NIST SP 800-63B, Digital Identity Guidelines. Why phone-network delivery is
RESTRICTED, and what a provider must do to keep using it. - NIST SP 800-63B supplement 1, Incorporating Syncable Authenticators, April 2024. When a synced passkey is acceptable at AAL2, and what the sync fabric has to do to get there.
- CISA, Implementing Phishing-Resistant MFA, October 2022. Which MFA forms fall to phishing, push bombing, SS7 and SIM swap.
- W3C Web Authentication Level 2. The relying party id and origin checks that make a lookalike domain unable to get a signature, and the difference between user presence and user verification.
- Google Chrome Help, Get your bookmarks, passwords and more on all your devices. What the optional sync passphrase changes, including whether saved passwords can be read at
passwords.google.com. - Yubico, Can I duplicate or back up a YubiKey?. Why there’s no copy to restore, and why the answer is a second registered key.
- GitHub, Recovering your account if you lose your 2FA credentials. One service’s recovery design, where a password reset does not bypass the second factor.
Since we recorded
NIST replaced the revision quoted above: SP 800-63B-4 superseded it on 1 August 2025 and absorbed the syncable-authenticator supplement, with the phone-network restriction intact. Microsoft added passkey saving and syncing to Microsoft Password Manager in Edge on 3 November 2025, joining Apple and Google on that list. The FIDO Alliance published its Credential Exchange Format as a Proposed Standard on 14 August 2025, a first step toward moving passkeys between providers. In Entra ID, Microsoft began a rollout on 1 September 2026 that makes passkeys the default sign-in method and prompts users enabled for SMS or voice to register one.
