Attackers can’t steal passkeys, so they enroll their own

Insights

Attackers can’t steal passkeys, so they enroll their own

Attackers can’t steal passkeys, so they enroll their own

Passkeys stopped password phishing at sign-in, so attackers now call employees and add a passkey of their own. Enrollment and recovery set your real security level.

Chad Spensky

·

A person holding a smartphone to their ear during a call, against a dark background

Passkeys took the stolen password off the table at sign-in. Attackers have moved to the moment a new passkey gets added, and to the way people get back in when they lose one.

Passkeys are winning at sign-in

Passkeys work. Across 15 large consumer services in the FIDO Alliance’s October 2026 Passkey Index, 36% of successful sign-ins now use a passkey, up from 26% in 2025. Passkey sign-ins succeed 93% of the time, against 78% for passwords and traditional MFA.

A passkey is bound to the real website. A look-alike domain gets nothing it can replay, so the attacker who used to phish a password and a one-time code has nothing to collect at sign-in. That attacker didn’t give up. They moved.

The phone call that enrolls the attacker

In July, Okta Threat Intelligence described a cluster it tracks as O-UNC-066 (Palo Alto Networks Unit 42 calls it CL-CRI-1147) going after the passkey enrollment process for Microsoft 365 customers. Okta saw targeting in food and beverage, technology, healthcare, automotive, construction and aviation, and says the main motive is data extortion.

The playbook is short:

  • The call. Someone phones an employee and says they need to register a new passkey, then sends them to a domain built around the word “passkey,” with a subdomain for each target.

  • The page. It copies the Microsoft sign-in flow, with the victim organization’s own logo and background.

  • The operator. A person watches a control panel that checks the victim’s browser every second and picks the next page: the password, then whichever second factor the account uses, whether an SMS code, an authenticator code or a push with number matching.

  • The swap. The victim believes they are creating a passkey. The real enrollment happens on the attacker’s side, directly with Microsoft, using what the victim typed.

Okta points out the timing. Since May 2026, Microsoft administrators can run passkey registration campaigns that nudge users at sign-in, and in some circumstances those nudges are on by default. A call saying “you need to set up a passkey” sounds like the rollout email everyone just received.

Microsoft has described the same pattern from its side. In intrusions it has tracked since May 2026, the attacker’s first goal after getting in was to add an MFA method under their control, “typically by registering a new phone number, authenticator application, or software-based one-time password (OTP) token,” so later sign-in challenges could be met without the user. The pretext, as The Hacker News summarized it, was a message posing as the help desk, urging employees to update their passkey, MFA or SSO settings to avoid losing access.

Five-step timeline of the enrollment attack: 1, a caller claiming to be IT says you need a new passkey; 2, a look-alike page with your company’s logo on a passkey-themed domain; 3, an operator relays your password and code to the real sign-in; 4, the attacker registers their own passkey on your account; 5, later sign-ins use the attacker’s passkey without you.
The passkey is never phished. The attacker adds one of their own while the victim thinks they are enrolling.

An account is as strong as the easiest way to add a key

None of this breaks passkeys. It uses the gap between two security levels on the same account. Microsoft’s documentation for registration campaigns says that before a passkey prompt, “the user completes MFA by using any method.” That is a sensible default for adoption. It also means a password and an SMS code are enough to reach the screen where a new passkey gets added.

If that is true for your accounts, their real security level is a password plus SMS, not a passkey. The same goes for the two other ways into an account that never touch the sign-in page: recovery, and a person at the help desk who can reset or add a factor.

Two stacked panels comparing three ways into the same account. Today: sign in needs a passkey, but adding a passkey needs only a password and an SMS code, and recovery needs only a help desk call or an emailed link. What it should require: a passkey to sign in, an existing passkey or a managed device to add one, and people you chose approving in their app to recover.
Sign-in is only one of three ways in. Adding a key and recovering access need the same strength.

Synced passkeys have a weaker door too

Enrollment is not the only way around a passkey. Many passkeys are synced: a cloud password manager stores them and copies them to every device signed into it. That makes a synced passkey only as strong as the password manager login, and that login can be phished. We showed this at DEF CON 33. A straightforward phishing page gets an attacker into the victim’s Chrome Password Manager, and with it the synced passkeys and passwords stored there. A similar attack against Bitwarden lets the attacker export the passkeys and import them into another password manager.

We picked Chrome because it is the most popular and Bitwarden because it is a trusted open-source option. To the best of our knowledge, every password manager that stores synced passkeys is open to the same attack, and an attacker inside the vault can also delete the keys and lock the victim out. The attack code is public.

The enrollment attack and this one have the same shape. Nobody breaks the passkey. The attacker adds a key of their own, or takes a copy of yours, through a weaker door. That is why Allthenticate passkeys are device-bound: the private key sits in your phone’s secure element, nothing is synced to a cloud account, and the phone has to be nearby to sign.

Two stacked panels. Synced passkey, stored in a cloud password manager: a phishing page copies the manager’s login, the attacker signs in to the password manager as you, reaches the synced vault of passkeys and passwords, and can export them or delete them and lock you out; our DEF CON 33 proof of concept does this against Chrome Password Manager and Bitwarden. Device-bound passkey, held in your phone as in Allthenticate: a sign-in request from the computer, the nearby phone signs with a key in its secure element, and returns a signed response for that site only. Nothing is synced to a cloud account, so there is no vault login to phish.
A synced passkey inherits the weakness of the login that guards its vault. A device-bound key has no vault to phish.

Close the enrollment path first

Okta’s recommendations and Microsoft’s detection guidance point the same way. In practice, that means four things:

  • Require a strong factor to add a strong factor. Registering a passkey should take an existing phishing-resistant credential or a managed device, not a password and a code.

  • Announce every authenticator change. Okta recommends notifying users of every factor lifecycle event. Send the same notice to your security team, and treat a new authenticator added minutes after a sign-in from an unfamiliar place as an incident until proven otherwise.

  • Give people a way to check the caller. Okta recommends establishing and communicating how users can verify help desk staff who contact them. The simplest rule is that a passkey is always set up by the user, from a known company page, never during an inbound call.

  • Retire the weak factors. Once someone has a passkey, remove SMS and other phishable methods from sign-in and from recovery. Otherwise the passkey guards the front door while the side door stays open.

Recovery that a caller can’t talk their way into

Recovery is the last door, and it usually ends in a help desk script or an emailed link. We built ours differently. In Allthenticate, your keys live in your phone, and you choose a rescue team: people you know, who accept your invitation in their own app. Your encrypted backup is kept for that team, and if you remove someone, it is re-encrypted for the rest.

When you lose your phone, you start a rescue mission from the new one and record one short video that includes the challenge the app shows you. Your rescuers see it in their app and approve, and the new phone gets your keys back, including your SSH keys. The recovery path has no password to reset and no help desk script to follow.

It isn’t magic. Your rescuers still have to pay attention. But an attacker now has to fool people who know your face and voice, inside an app that shows them exactly what they are approving, instead of fooling one stranger reading from a script.

The takeaway

Rolling out passkeys moves the attack. It doesn’t end it. Before your next registration campaign goes out, check what it takes to add a passkey to one of your accounts, and what it takes to recover one. If either answer is a password and a code, that is your real security level. If your passkeys sync through a password manager, check that login too.

If you want to see a phone that holds your keys, with recovery through people you trust, download Allthenticate or contact us.

Passkeys took the stolen password off the table at sign-in. Attackers have moved to the moment a new passkey gets added, and to the way people get back in when they lose one.

Passkeys are winning at sign-in

Passkeys work. Across 15 large consumer services in the FIDO Alliance’s October 2026 Passkey Index, 36% of successful sign-ins now use a passkey, up from 26% in 2025. Passkey sign-ins succeed 93% of the time, against 78% for passwords and traditional MFA.

A passkey is bound to the real website. A look-alike domain gets nothing it can replay, so the attacker who used to phish a password and a one-time code has nothing to collect at sign-in. That attacker didn’t give up. They moved.

The phone call that enrolls the attacker

In July, Okta Threat Intelligence described a cluster it tracks as O-UNC-066 (Palo Alto Networks Unit 42 calls it CL-CRI-1147) going after the passkey enrollment process for Microsoft 365 customers. Okta saw targeting in food and beverage, technology, healthcare, automotive, construction and aviation, and says the main motive is data extortion.

The playbook is short:

  • The call. Someone phones an employee and says they need to register a new passkey, then sends them to a domain built around the word “passkey,” with a subdomain for each target.

  • The page. It copies the Microsoft sign-in flow, with the victim organization’s own logo and background.

  • The operator. A person watches a control panel that checks the victim’s browser every second and picks the next page: the password, then whichever second factor the account uses, whether an SMS code, an authenticator code or a push with number matching.

  • The swap. The victim believes they are creating a passkey. The real enrollment happens on the attacker’s side, directly with Microsoft, using what the victim typed.

Okta points out the timing. Since May 2026, Microsoft administrators can run passkey registration campaigns that nudge users at sign-in, and in some circumstances those nudges are on by default. A call saying “you need to set up a passkey” sounds like the rollout email everyone just received.

Microsoft has described the same pattern from its side. In intrusions it has tracked since May 2026, the attacker’s first goal after getting in was to add an MFA method under their control, “typically by registering a new phone number, authenticator application, or software-based one-time password (OTP) token,” so later sign-in challenges could be met without the user. The pretext, as The Hacker News summarized it, was a message posing as the help desk, urging employees to update their passkey, MFA or SSO settings to avoid losing access.

Five-step timeline of the enrollment attack: 1, a caller claiming to be IT says you need a new passkey; 2, a look-alike page with your company’s logo on a passkey-themed domain; 3, an operator relays your password and code to the real sign-in; 4, the attacker registers their own passkey on your account; 5, later sign-ins use the attacker’s passkey without you.
The passkey is never phished. The attacker adds one of their own while the victim thinks they are enrolling.

An account is as strong as the easiest way to add a key

None of this breaks passkeys. It uses the gap between two security levels on the same account. Microsoft’s documentation for registration campaigns says that before a passkey prompt, “the user completes MFA by using any method.” That is a sensible default for adoption. It also means a password and an SMS code are enough to reach the screen where a new passkey gets added.

If that is true for your accounts, their real security level is a password plus SMS, not a passkey. The same goes for the two other ways into an account that never touch the sign-in page: recovery, and a person at the help desk who can reset or add a factor.

Two stacked panels comparing three ways into the same account. Today: sign in needs a passkey, but adding a passkey needs only a password and an SMS code, and recovery needs only a help desk call or an emailed link. What it should require: a passkey to sign in, an existing passkey or a managed device to add one, and people you chose approving in their app to recover.
Sign-in is only one of three ways in. Adding a key and recovering access need the same strength.

Synced passkeys have a weaker door too

Enrollment is not the only way around a passkey. Many passkeys are synced: a cloud password manager stores them and copies them to every device signed into it. That makes a synced passkey only as strong as the password manager login, and that login can be phished. We showed this at DEF CON 33. A straightforward phishing page gets an attacker into the victim’s Chrome Password Manager, and with it the synced passkeys and passwords stored there. A similar attack against Bitwarden lets the attacker export the passkeys and import them into another password manager.

We picked Chrome because it is the most popular and Bitwarden because it is a trusted open-source option. To the best of our knowledge, every password manager that stores synced passkeys is open to the same attack, and an attacker inside the vault can also delete the keys and lock the victim out. The attack code is public.

The enrollment attack and this one have the same shape. Nobody breaks the passkey. The attacker adds a key of their own, or takes a copy of yours, through a weaker door. That is why Allthenticate passkeys are device-bound: the private key sits in your phone’s secure element, nothing is synced to a cloud account, and the phone has to be nearby to sign.

Two stacked panels. Synced passkey, stored in a cloud password manager: a phishing page copies the manager’s login, the attacker signs in to the password manager as you, reaches the synced vault of passkeys and passwords, and can export them or delete them and lock you out; our DEF CON 33 proof of concept does this against Chrome Password Manager and Bitwarden. Device-bound passkey, held in your phone as in Allthenticate: a sign-in request from the computer, the nearby phone signs with a key in its secure element, and returns a signed response for that site only. Nothing is synced to a cloud account, so there is no vault login to phish.
A synced passkey inherits the weakness of the login that guards its vault. A device-bound key has no vault to phish.

Close the enrollment path first

Okta’s recommendations and Microsoft’s detection guidance point the same way. In practice, that means four things:

  • Require a strong factor to add a strong factor. Registering a passkey should take an existing phishing-resistant credential or a managed device, not a password and a code.

  • Announce every authenticator change. Okta recommends notifying users of every factor lifecycle event. Send the same notice to your security team, and treat a new authenticator added minutes after a sign-in from an unfamiliar place as an incident until proven otherwise.

  • Give people a way to check the caller. Okta recommends establishing and communicating how users can verify help desk staff who contact them. The simplest rule is that a passkey is always set up by the user, from a known company page, never during an inbound call.

  • Retire the weak factors. Once someone has a passkey, remove SMS and other phishable methods from sign-in and from recovery. Otherwise the passkey guards the front door while the side door stays open.

Recovery that a caller can’t talk their way into

Recovery is the last door, and it usually ends in a help desk script or an emailed link. We built ours differently. In Allthenticate, your keys live in your phone, and you choose a rescue team: people you know, who accept your invitation in their own app. Your encrypted backup is kept for that team, and if you remove someone, it is re-encrypted for the rest.

When you lose your phone, you start a rescue mission from the new one and record one short video that includes the challenge the app shows you. Your rescuers see it in their app and approve, and the new phone gets your keys back, including your SSH keys. The recovery path has no password to reset and no help desk script to follow.

It isn’t magic. Your rescuers still have to pay attention. But an attacker now has to fool people who know your face and voice, inside an app that shows them exactly what they are approving, instead of fooling one stranger reading from a script.

The takeaway

Rolling out passkeys moves the attack. It doesn’t end it. Before your next registration campaign goes out, check what it takes to add a passkey to one of your accounts, and what it takes to recover one. If either answer is a password and a code, that is your real security level. If your passkeys sync through a password manager, check that login too.

If you want to see a phone that holds your keys, with recovery through people you trust, download Allthenticate or contact us.

Your phone is the key

Passwordless login for desktops, SSH, SSO and doors, with proximity lock when you walk away.

Download Allthenticate