people in server room

FIDO Didn't Break. Microsoft's Analysis Shows the Flows Around It Did.

On September 9, 2026, Microsoft published its analysis of a cloud intrusion campaign that has been active since May 2026. The security press headlined it immediately as “passkey phishing.”

That shorthand is misleading, and it carries a real cost. It suggests that an organization that has deployed passkeys remains exposed at the same level as before. Taken literally, that framing could slow passwordless programs at the exact moment they should be accelerating.

In this campaign, no Fast IDentity Online (FIDO) private key was stolen. No signature was replayed. The word “passkey” appears in the social engineering script, not in the cryptography.

The campaign highlights two weaknesses: unsafe authentication flows before sign-in and insufficiently protected authenticator enrollment after sign-in. Here is exactly how, and what it means for your roadmap.

What the Microsoft Report Actually Describes

The attack unfolds in three stages: pretext, capture and persistence.

1. The pretext. The actor calls or texts the employee’s personal phone number, posing as the internal IT help desk. The message creates a sense of urgency: your passkey, multi-factor authentication (MFA), or single sign-on (SSO) configuration must be updated immediately, or your access will be disrupted.

2. The capture. Microsoft describes two variants, and the distinction matters:

  • In an adversary-in-the-middle (AitM) flow, in which the user is redirected to a counterfeit site mimicking the Microsoft sign-in page, and the proxy intercepts credentials and the session cookie
  • In a device code flow, the attacker initiates authentication from their own client and guides the victim into approving it. No credentials need to be stolen. The victim completes authentication on the legitimate Microsoft service, but the resulting token is issued to the attacker-controlled client.

3. The persistence. Early in the intrusion, the actor establishes persistence by registering an authentication method under their own control: a new phone number, an authenticator application, or a software-based one-time password  (OTP) token. A temporary compromise becomes a persistent foothold. The attacker-controlled method remains registered until defenders explicitly remove it, and remediation must also revoke compromised sessions and tokens.

From there, activity expands to internal reconnaissance via the Graph API, inventorying users, groups and permissions, followed by sustained exfiltration from SharePoint Online, OneDrive and Exchange Online over hours or days.

The domains registered by the actor confirm the nature of the lure: passkey help desk, secure-passkey, set up my passkey, add-passkey, and integrated SSO. The vocabulary of modern security has itself become a social engineering asset.

Image
The Attack in One Sentence Attackers did not defeat FIDO authentication. They used social engineering to gain initial access and then established persistence by enrolling authentication methods under their own control.

What FIDO2 Protects, and Where Governance Must Extend Further

Three properties of the FIDO2/WebAuthn standard are nowhere challenged in the report:

  • The private key is never disclosed to the relying party or to a phishing site. There is no shared secret to intercept, no code to relay. With a device-bound hardware authenticator, the credential also remains on the authenticator and cannot be synchronized to another device.
  • The assertion is bound to the origin. The browser includes the actual origin in the signed data. A portal hosted on a look-alike domain cannot obtain a valid signature for the legitimate one. The AitM proxy, devastatingly effective against a one-time code or a push notification, comes away with nothing usable.
  • Each authentication ceremony uses a fresh, unpredictable challenge. A captured assertion cannot be replayed as a new authentication.

Against the credential-capture mechanism used by the AitM variant, a user who authenticates exclusively via a correctly implemented FIDO2 flow is protected. The counterfeit site cannot obtain a valid assertion for the legitimate origin.

The device code variant is different, and this is the point most coverage has missed. That protection applies when the attacker impersonates the relying party. In device code phishing, the victim may authenticate with the legitimate relying party using a perfectly valid passkey while authorizing an attacker-controlled client. WebAuthn works exactly as designed. The user simply approves the wrong authentication context.

Image
The lesson for CISOs is precise: FIDO makes the credentials phishing-resistant. It does not automatically make every authentication or authorization flow phishing-resistant. Microsoft now classifies device code flow as a high-risk authentication flow and recommends blocking it wherever possible. Phishing-resistant MFA and the restriction of unsafe flows are two separate controls, and both are needed.

The Larger Governance Gap: Enrollment and Recovery

Once initial access is obtained through either variant, one of the actor’s first objectives is to add an authentication factor under their own control. This is where most organizations have a second gap.

A FIDO deployment in which an emailed link can bootstrap enrollment without fresh strong authentication, and in which recovery relies on a call to the help desk, has hardened sign-in and left the registration of new methods untouched. The attacker does not break down the door. They get handed a second key.

Six Measures to Strengthen the Full Authentication Journey

  • Remove phishable fallback factors rather than merely deprioritizing them. If a phishable fallback, such as an SMS code or a context-free push notification, can satisfy the same access policy as a passkey, then that weaker path determines the policy’s effective phishing resistance. The attacker will always pick it.
  • Block the device code flow wherever no documented business requirement exists. This flow was designed for input-constrained devices. Where no such requirement exists, it should be blocked rather than enabled tenant-wide, with exceptions documented and scoped.
  • Treat enrollment as a high-sensitivity operation. Enroll in person or derive the new factor from an existing strong one, or issue a single-use, short-lived temporary access credential delivered out of band. Never allow possession of an emailed link, by itself, to authorize enrollment.
  • Restrict the registration of new authentication methods. The authority to create a new authenticator is itself an authentication decision. It should require assurance at least as strong as the credential being enrolled: fresh interactive authentication using a phishing-resistant method, reinforced where appropriate by a managed device, named location and risk-based controls. This is the direct countermeasure to the observed persistence mechanism.
  • Rebuild the help desk procedure. The attack impersonates support, so the response must be symmetrical: no factor reset on an inbound call, documented identity verification, systematic callback on an official channel and clear internal communication stating that support never asks anyone to update a passkey over the phone.
  • Monitor enrollment, not just sign-in. Microsoft makes the point directly: a single Graph API call rarely looks suspicious in isolation. Detection has to follow behavioral progression. High-value detection comes from correlating an unusual sign-in with authentication-method enrollment, followed by Graph reconnaissance or abnormal cloud access.

How Governance Controls Address the Attack

Governance ControlRisk Addressed
Remove phishable fallback factorsCredential phishing paths
Restrict device code flowAttacker-initiated session authorization
Secure enrollmentUnauthorized authenticator registration
Restrict new authentication methodsAttacker persistence
Strengthen help desk proceduresSocial engineering attacks
Monitor enrollment eventsPost-compromise activity

From Deployed Passkeys to Governed Passkeys

These measures share a common logic: regaining control at the moment an authenticator enters the system. Four layers of assurance build on each other here, and it helps to keep them distinct.

The FIDO standard provides phishing resistance. A device-bound hardware authenticator ensures that the credential is not synchronizable across devices. Attestation provides provenance information about the authenticator model being registered. And Enterprise Attestation, combined with a controlled issuance process, turns provenance into governance by enabling the relying party to obtain stronger, organization-specific information about the authenticator at the time of registration.

Two precisions matter in practice. An Authenticator Attestation Globally Unique Identifier (AAGUID) identifies a make and model, not an individual device. When combined with verified attestation, an AAGUID-based policy can restrict enrollment to approved authenticator models; without enforced attestation, it remains a policy guide rather than a strict security control.

Governed passkeys require more than phishing-resistant authentication. They require trustworthy credential issuance, auditable enrollment and life cycle controls that help ensure authenticators remain tied to verified identities.

The HID Crescendo 4000 Cards and Crescendo Key authenticators are device-bound and support Enterprise Attestation. Beyond phishing resistance, they add protection against credential synchronization described above, and the convergence of physical badge and logical identity allows digital enrollment to be anchored to a badge-issuance process that already involves in-person verification.

Enrollment ceases to be an online form and becomes a regulated act.

Key Takeaways for Identity Leaders

The campaign illustrates a governance gap, not a failure of FIDO cryptography. It demonstrates that FIDO makes credentials phishing-resistant, but doesn’t automatically make enrollment, recovery, authorization flows or help desk procedures phishing-resistant on their own. Each part of the journey has to be brought to the same standard.

The question to put to your organization is not “have we deployed passkeys?” It is twofold: which authentication flows can still authorize an attacker-initiated session, and how many paths remain for an attacker to enroll a factor on our behalf.

Assess how your FIDO authenticators are issued, registered, recovered and monitored. HID can help organizations connect phishing-resistant authentication with controlled credential issuance and life cycle management. 

Frequently Asked Questions

1. If passkeys are phishing-resistant, how were attackers able to gain access?

The campaign described by Microsoft did not rely on stealing a FIDO private key or replaying a WebAuthn assertion. Instead, attackers used social engineering to manipulate users into completing unsafe authentication flows or authorizing attacker-controlled clients. The lesson is that phishing-resistant credentials must be paired with secure authentication and authorization flows.

2. Does this mean passkeys are no longer effective?

No. The protections provided by FIDO2 and WebAuthn remain unchanged. Passkeys help prevent credential theft and credential replay attacks. What this campaign demonstrates is that organizations must also secure enrollment, recovery and authentication workflows surrounding the credential.

3. What are governed passkeys?

Governed passkeys combine phishing-resistant authentication with controls for credential issuance, enrollment, recovery, replacement, revocation and life cycle management. The goal is to extend phishing resistance beyond sign-in and throughout the authentication journey.

4. Why is authentication-method enrollment such a critical control?

Once an attacker gains temporary access, one of the fastest paths to persistence is registering an authentication method they control. If enrollment and recovery processes are weaker than sign-in controls, an attacker may maintain access even after the initial compromise vector is removed.

5. What should organizations review first?

Start by identifying:

  • Authentication flows that can authorize attacker-initiated sessions
  • Phishable fallback factors that satisfy the same policies as passkeys
  • Enrollment and recovery processes that do not require fresh strong authentication
  • Help desk procedures that can reset or replace credentials with insufficient verification
6. What is the difference between deploying passkeys and governing passkeys?

Deploying passkeys improves phishing resistance at sign-in. Governing passkeys means controlling how authenticators are issued, registered, recovered, replaced and revoked throughout their life cycle. The strongest programs address both authentication strength and authenticator governance.