Pass-ta-key attacks show that even passkeys have an attack surface

Passkeys have increasingly been promoted as the future of authentication. They eliminate the need to remember a password, they’re resistant to traditional phishing attacks, and the private portion of a passkey isn’t supposed to leave the device or credential manager protecting it.

But researchers from Palo Alto Networks’ Unit 42 have demonstrated something important: attackers don’t necessarily have to break the cryptography behind a passkey if they can compromise the system surrounding it.

The research describes three attacks against Google Password Manager’s synchronized passkeys on Windows. The techniques have been given the names Pass-ta-key, Silver Pass-ta-key and Golden Pass-ta-key.

Before anyone decides that this means passkeys are useless, there is an important qualification.

The computer must already be compromised by malware.

The researchers aren’t describing someone remotely cracking a passkey simply because you use one. They’re demonstrating what malware already running on a Windows computer may be able to do with the browser, credential manager and synchronization infrastructure that handles those passkeys.

Unit 42’s original research describes the attacks in detail, while BleepingComputer covered the findings after the research became public.

Passkeys aren’t being cryptographically broken

This distinction is important.

The researchers aren’t claiming that they defeated the underlying WebAuthn cryptography. Instead, they’re attacking the infrastructure around it.

Think about everything involved when a passkey is used.

There is the device. There is the browser. There is the credential manager. There may be cloud synchronization between devices. There are mechanisms for enrolling and recovering devices. Finally, there is the website accepting the authentication request.

Every one of those pieces becomes part of the security equation.

If malware is already running on the computer, an attacker may not need to crack the passkey itself. They may instead manipulate one of the systems trusted to use it.

The first attack: Pass-ta-key

The basic Pass-ta-key attack allows malware running without administrator privileges to impersonate a trusted device and request a valid authentication response from Google Password Manager.

According to Unit 42, Chrome on Windows uses a hardware-backed device identity key. Although the key is protected, information associated with it can be accessed by malware running under the user’s account and used through Windows cryptographic functions.

The result is particularly interesting.

Malware may be able to obtain a valid authentication assertion without the person actually approving the authentication.

Whether that assertion succeeds depends partly on the website receiving it.

And that’s where this research gets even more interesting.

GitHub said no. eBay said yes.

Unit 42 tested the technique against multiple services.

GitHub rejected the authentication attempt because it checked whether user verification had actually occurred.

When the researchers tested eBay, however, the authentication succeeded because the service wasn’t properly validating the WebAuthn user-verification flag.

That difference demonstrates why implementing passkeys isn’t simply a matter of adding a button that says ‘Sign in with a passkey.’

The website accepting the credential has responsibilities too.

After Unit 42 reported the problem, eBay corrected its implementation.

This is an important lesson for companies deploying passwordless authentication: requiring user verification and actually checking that verification occurred are not necessarily the same thing.

Silver Pass-ta-key goes further

The researchers then developed a more powerful technique they call Silver Pass-ta-key.

Instead of merely obtaining an authentication response, this technique targets the process used when a device is registered or re-registered with Google Password Manager.

The attacker can manipulate that process so that an attacker-controlled verification key becomes trusted.

That changes the situation considerably.

Once the attacker’s key has been registered, the attacker may be able to request verified authentication assertions from another computer. The original compromised machine doesn’t necessarily need to remain available.

In other words, what began as malware running locally could potentially become persistent access from elsewhere.

Device enrollment and recovery exist because people buy new computers, lose devices and otherwise need legitimate ways to regain access to their credentials.

But recovery mechanisms have always presented an interesting security problem.

If you build a way for the legitimate owner to recover access, an attacker is going to be interested in whether that same mechanism can be abused.

Passkeys don’t make that problem disappear.

Then comes Golden Pass-ta-key

The third technique is potentially the most serious.

Unit 42 calls it Golden Pass-ta-key.

Google Password Manager can synchronize passkeys between devices. Those passkeys are encrypted, and a security-domain secret is involved in protecting the synchronized credentials.

The researchers found that this secret could be exposed.

Originally, Unit 42 found sensitive information appearing in Chrome’s FIDO-related logging. Google changed Chrome so that the secret is no longer written to those logs after the researchers disclosed the issue.

But that didn’t completely end the problem.

According to Unit 42, the security-domain secret still temporarily exists in Chrome’s process memory during certain operations, such as re-registering the device with Google’s cloud authenticator.

Malware running as the user could potentially obtain that secret from memory.

If an attacker obtains the security-domain secret and the encrypted passkey database, the researchers demonstrated that synchronized passkey private keys could be decrypted.

That is a very different problem from stealing one authentication session.

The researchers also raised another concern: there currently isn’t a straightforward mechanism for a user to rotate or revoke the compromised security-domain secret.

That could potentially affect passkeys synchronized in the future as well.

Passkeys themselves have been discussed extensively over the years. Steve Gibson of Security Now has covered them numerous times, including Security Now episode 870, ‘That Passkeys Thing,’ and later episode 965, ‘Passkeys vs 2FA.’ Gibson has argued that passkeys provide significant security advantages over passwords and traditional two-factor authentication. The Pass-ta-key research doesn’t necessarily contradict that. Instead, it demonstrates that even a strong authentication technology depends on the security of the browser, credential manager, operating system and website implementations surrounding it.

This doesn’t mean you should stop using passkeys

It would be very easy to read a headline about malware stealing passkeys and conclude that the entire passwordless authentication idea has failed.

That isn’t what the researchers are saying.

Unit 42 continues to recommend passkeys and describes them as significantly safer than passwords.

Passkeys still eliminate several major problems associated with passwords.

They aren’t reusable secrets that someone can simply type into a phishing website. A server breach doesn’t expose a password that can then be tried elsewhere. Users don’t have to invent memorable passwords or reuse them between services.

Those are significant improvements.

What this research demonstrates is that better authentication doesn’t make endpoint security irrelevant.

If malware is already running as you on your computer, the attacker has crossed an extremely important security boundary.

Malware remains the starting point

All three attacks described by Unit 42 begin with an already-compromised Windows machine.

That makes the usual security advice just as relevant as it was before passkeys existed.

• Keep the operating system and browser updated.
• Keep security software and other protections enabled.
• Be cautious about programs and files you download and execute.
• Be suspicious of unexpected attachments and links.
• Don’t assume passwordless authentication protects a compromised computer.

For organizations implementing passkeys, there are additional lessons.

Websites need to properly require and validate user verification. Credential managers need to carefully protect enrollment and recovery processes. Sensitive cryptographic material should be protected not only while stored on disk but also while being processed in memory.

Those aren’t problems an ordinary user can solve.

They are implementation problems for browser developers, credential providers and websites.

The attack is moving around the credential

For years, attackers went after passwords because passwords were the easiest thing to steal.

Phishing pages asked people to type them in. Malware extracted them from browsers. Credential-stealing attacks collected enormous databases of usernames and passwords.

Passkeys remove many of those opportunities.

So attackers will look elsewhere.

The Pass-ta-key research is a good example of what that future may look like.

Instead of asking, ‘How do I steal the password?’ an attacker can ask different questions.

Can I impersonate a trusted device?

Can I interfere with device enrollment?

Can I abuse account recovery?

Can I convince a website that user verification occurred when it didn’t?

Can I steal the material used by a credential manager to synchronize credentials?

Those questions don’t require breaking the mathematical foundation of passkeys.

They attack the systems surrounding them.

That’s an important distinction, because we’re likely to see more research like this as passkeys become more widely deployed.

Passkeys are still a major improvement over passwords.

But as Pass-ta-key, Silver Pass-ta-key and Golden Pass-ta-key demonstrate, removing the password doesn’t remove the attacker.

It just changes what the attacker goes after.