# Passkeys: the password that doesn't exist

> If the trouble with passwords is that there's a shared secret someone else keeps on your behalf, the fix is to get rid of the shared secret. That's what passkeys are — here's what they fix and what they don't.

2026-06-18 · David Carrero · password.es
Original: https://password.es/en/blog/passkeys-the-password-that-doesnt-exist/

---

Every password story ends the same way: somebody was holding your secret
somewhere, and they lost it.

Pick any year you like. In the beginning it sat in a plain text file on a shared
machine, and all it took was asking the system to print it. In 1979, Morris and
Thompson wrote up how to salt the hash and make computing it expensive, so that a
stolen list didn't instantly become a list of passwords — an enormous improvement.
Then came the composition rules that had us all bolting a symbol onto the end. But
the shape of the problem didn't budge an inch in all that time: **there is a
secret, you know it, the server can check it, and therefore the server has
something to lose**.

Everything we've done since has been about making that secret longer, weirder,
saltier, slower to crack. Nobody asked the question that comes before all that:
what if there were no shared secret at all?

## 1976: the idea that a secret can have two halves

In 1976, Whitfield Diffie and Martin Hellman published *New Directions in
Cryptography*. The idea was almost insulting in its simplicity: **a key can be
split into two pieces that aren't interchangeable**. One you hand out to anyone
who asks; the other never leaves the house. What you sign with the private half is
verified with the public half, but the public half can't sign a thing.

Read that again with a server in mind: the site no longer needs to keep anything
that can be used to get in. Only the half that opens nothing.

That it took the better part of fifty years for this to reach your bank's login
screen doesn't say much for our industry. Client certificates existed and they
worked; what didn't exist was a way to use them that didn't require a graduate
degree. The cryptography was solved. The usability wasn't.

## 2013: for once, somebody agrees on something

The **FIDO Alliance** was founded in 2013, an industry consortium with a stated
and not especially modest goal: reduce the world's reliance on passwords. What
mattered wasn't the technology — that part was already done — but getting device
makers, browsers and services around one table to agree on a single protocol. It
sounds like committee work and it was exactly what was missing: an authentication
mechanism that only works in one browser with one service isn't a mechanism, it's
a demo.

## 2019: the browser learns cryptography

In **March 2019, WebAuthn became a W3C Recommendation**. In plain English: any web
page could now ask the browser — through a standard API, no plugins, no drivers —
to generate a key pair and sign with it.

The flow, stripped of decoration, goes like this. When you register, your device
creates a fresh key pair specific to that site. It sends the public half. It keeps
the private half, held by the device itself and protected by your fingerprint,
your face or its PIN. When you come back, the site sends a random challenge, your
device signs it, and the site checks the signature against the public key it
already had.

Notice what **doesn't** happen: no secret travels anywhere. The server never had
your private key, so it can't lose it. If its entire database leaks tomorrow, what
walks out the door is a list of public keys — roughly as valuable as stealing a
phone book. No hashes to run through a GPU, no dictionary to try. It isn't that
it's hard: it's that there's nothing there worth having.

## The side effect is bigger than the idea

When your device creates that key pair, it binds it to the domain. The key it
saved for `yourbank.com` **is only ever offered to `yourbank.com`**. This isn't a
policy, or a warning, or a red banner in the address bar: the key for
`yourbank-security.com` simply doesn't exist. The browser can't find it because it
was never created.

This kills phishing. Not reduces it — leaves it without a mechanism. The entire
phishing business rests on the fact that **you can type your password into the
wrong site**, and we've spent years trying to patch that with user training and
awareness campaigns. All of it treated the symptom. Your eye can be fooled by a
lookalike URL; your phone can't. It doesn't read. It compares strings.

It takes reuse down with it, which was the other cardinal sin. Every site gets its
own key pair. There's nothing to reuse even if you wanted to.

## 2022: the big three sign a peace treaty

In **May 2022, Apple, Google and Microsoft jointly announced** that they would
expand support for FIDO credentials — passkeys — across their platforms and
browsers.

That's what turned this from a technical footnote into something you can explain
to your mother. Until then, WebAuthn nearly always meant a physical USB key:
excellent, and used by about three people. A passkey lives where your phone
already is — in your hand — and unlocks with your face.

## What it doesn't fix

Now the small print, because the industry is going to spend the next few years
pretending it isn't there.

**Recovery is still the soft spot.** The hole hasn't gone away: it's moved. Lose
the device and somebody has to decide that you're you, and that decision is nearly
always made over email, over SMS, or by talking to a human in a call centre. That
human is exactly the same target they always were. You've armoured the front door
and the back one is right where you left it.

**Syncing ties you to an ecosystem.** Passkeys are convenient because your
platform replicates them across your devices. Which is another way of saying your
keys live on one particular company's keyring. Getting out isn't impossible, but
it isn't a button either. You traded the problem of remembering for the problem of
belonging.

**And you still need a way in when you lose the phone.** Which in practice means
almost no site has actually removed the password: it has tucked it behind a
"trouble signing in?" link. As long as that link exists, the password exists, and
it's as bad as it ever was. The chain breaks at the weakest method you'll accept,
not the most elegant one you offer.

## What to do in the meantime

Turn on passkeys wherever they're offered, and start with whatever hurts most if
it falls: your email, your bank, the account you use to log into everything else.
It's one of the few security improvements that also saves you work.

And don't kid yourself: you'll be living with passwords for years yet, if only as
a plan B. Make them long, make them different everywhere, and keep them in a
manager. If you want one that never leaves your head, use the
[generator](/en/); if you want to know whether the one you've got holds up at
all, run it through the [checker](/en/checker/).

The old sentry's question — *are you one of us?* — finally has a different answer.
For decades we answered by saying a word that could be overheard. Now we answer
with a signature only your device can produce, without the word ever coming into
existence. It's the first time we've changed the question instead of the word.

---

*Sources: W. Diffie and M. Hellman, "New Directions in Cryptography", IEEE
Transactions on Information Theory, 1976 · R. Morris and K. Thompson, "Password
Security: A Case History", Communications of the ACM, 1979 · FIDO Alliance,
founded in 2013 · "Web Authentication: An API for accessing Public Key
Credentials", W3C Recommendation, March 2019 · joint announcement by Apple, Google
and Microsoft on passkey support, May 2022.*