One Allow tap on a Google Workspace Super Admin push handed Russia-linked COLD RELIC a tenant-wide seat. According to Google Threat Intelligence Group and Mandiant, the operators signed in with stolen Super Admin credentials, repeated MFA push notifications until one was accepted, then later phished Microsoft Entra ID credentials and authentication tokens used for federated access into a second US organization's Google Workspace. Phishing-proof device-bound MFA removes spam-approvable pushes and live-stealable login factors at those gates, so neither Super Admin fatigue nor the Entra capture completes sign-in in the first place.

Why Super Admin push and Entra federation still transferred

COLD RELIC authenticated to a US organization's Google Workspace Super Admin account with stolen credentials and a multi-factor authentication push notification fatigue attack. GTIG and Mandiant state that the operators routed the login through a US residential proxy geolocated near the victim. After the push was accepted, they registered a previously unseen device model, rogue MFA devices, backup codes, and passkeys, then used Super Admin rights for domain-wide Customer Takeout export, mass Drive sharing and downloads, employee roster export, admin log queries, and deletion of Alert Center notifications.

Stolen Super Admin passwords only matter if the second factor can travel with them. Push-approve MFA is transferable attention. Anyone who can trigger prompts can wait for a hurried admin to tap Allow. COLD RELIC did exactly that on the highest-privilege Google Workspace identity, with a residential proxy reducing the chance a location anomaly alone blocked the sign-in. The approved push produced a normal Super Admin session. No origin-bound private key ceremony stood between the stolen password and tenant-wide power.

Persistence after that session was enrollment abuse, not a fresh login puzzle. GTIG and Mandiant describe attacker-enrolled passkeys that later supported reauthentication across password and session resets and bypassed device registration controls. Backup codes and rogue MFA devices sat beside OAuth inbox sync to external clients. Password change alone did not evict operators who already owned factors the IdP treated as legitimate. That failure matches coaching a user through a fake login: a transferable enrollment or recovery path lets the attacker become an authenticator owner.

Several months later the same crew impersonated an employee of that first victim over encrypted messaging and phished a second US organization that used Microsoft Entra ID as the federated identity provider for Google Workspace. Phishing links collected Entra ID credentials and authentication tokens, then redirected victims to document lures tied to the ongoing chat. After capture, operators registered a new MFA device in the identity stack and bulk-downloaded cloud Drive files. Federation made Entra ID the front door to Google Workspace. One IdP proof unlocked Workspace without a separate Google-native ceremony.

Public reporting does not name the victims or quantify exfil volumes. The full attack narrative lives on the companion write-up: COLD RELIC Super Admin MFA push fatigue explained.

Stage What transferred Login MFA posture
Super Admin sign-in Password plus push approval Fatigueable second factor
Post-login enrollment Passkeys, codes, rogue devices Session trusted for new factors
Entra federated path Credentials and auth tokens Phishable IdP login
After password reset Unrevoked sessions and planted factors Reset without full revoke

Device-bound proof across login and enrollment

Device-bound phishing-proof authentication changes the Super Admin gate first. There is no push to spam, no OTP to type, and no approval prompt an attacker can fatigue from a residential proxy. The private key stays in hardware on an already-enrolled device. The server holds only public keys. A remote operator with a stolen password cannot complete a signature that never leaves the victim's TPM or secure enclave. Super Admin push fatigue as GTIG and Mandiant documented it simply has no factor left to hammer.

The same bound key stops the Entra path that fed federated Google Workspace. Live capture of passwords and authentication tokens needs a login ceremony the attacker can finish or proxy. Origin-bound, device-bound signatures refuse a wrong origin and never put a reusable secret on the wire. The second victim's federated front door never issues the session the phishing kit was built to steal. That is prevention, not detection: the campaign fails before a Super Admin or Entra session exists.

Device-bound keys can close those login gates. That is not enough for the persistence GTIG and Mandiant documented. After the stolen Super Admin session existed, COLD RELIC enrolled passkeys, backup codes, and rogue MFA devices that later survived password and session resets and bypassed device registration controls. The second tenant repeated the pattern: after Entra capture, operators registered a new MFA device in the identity stack.

If enrollment stays reachable from a hijacked Super Admin or IdP session, a strong login factor can still be planted by the attacker. The fix is to require an already-enrolled device before a new factor is added, then revoke sessions, OAuth grants, and attacker-enrolled authenticators together. That enrolment model is what stops a planted passkey from becoming the durable identity after the first push approval.

If an organization hardened only the login form and left Super Admin or recovery able to enroll new factors from a compromised session, COLD RELIC-style persistence would still land.

After the session already exists

Domain-wide Customer Takeout, mass Drive downloads, roster export, and Alert Center deletion are authorization abuse after a privileged session already exists. Login cryptography does not unwind admin APIs. Stolen Entra authentication tokens and unrevoked attacker sessions after a completed sign-in are token-stage problems. Revoke access and refresh tokens, kill attacker-enrolled factors, and force re-proof on every device. Shortening token lifetime alone does not prevent the original capture; it only narrows the replay window if the attacker never planted a lasting factor.

Control Super Admin push fatigue Entra credential and token phishing After session exists
Push or OTP MFA Fatigued or relayed Captured live Does not undo admin APIs
Passkeys at login only Strong at form Strong at form Enrollment may still open
Phishing-proof MFA 2.0 No push factor No transferable login secret Does not undo admin APIs; revoke sessions and planted factors

Key takeaways for defenders

Privileged cloud identity will keep drawing state-linked operators. The durable answer is a Super Admin and federated IdP path that never presents a push to fatigue or a secret to relay, and never lets a stolen session mint lasting authenticators.

FAQ

Would phishing-proof MFA have stopped COLD RELIC Super Admin push fatigue?

Yes. Phishing-proof device-bound MFA would have stopped the COLD RELIC Super Admin push fatigue path because there is no spam-approvable push or other transferable second factor for operators to hammer after they hold a stolen password. According to Google Threat Intelligence Group and Mandiant, COLD RELIC authenticated with stolen Super Admin credentials plus repeated MFA push notifications until one was accepted; a hardware-bound signature that never leaves the enrolled device gives that campaign nothing to approve remotely.

Would passkeys alone have blocked attacker persistence in the COLD RELIC case?

Passkeys alone would not necessarily have blocked COLD RELIC persistence if enrollment stayed reachable from a hijacked Super Admin session. GTIG and Mandiant describe attacker-enrolled passkeys that persisted across password and session resets and bypassed device registration controls. With MFA 2.0 as the only sign-in and enrollment method, a stolen session token is only that login. Once the token expires it cannot be reused, and the attacker cannot add a device.

Would phishing-proof MFA have stopped the Entra path into the second Google Workspace tenant?

It would have stopped the Entra login if that path still depended on phishable credentials or other transferable login secrets. GTIG and Mandiant say the second US organization used Microsoft Entra ID as the federated identity provider for Google Workspace, and the phishing links were designed to steal Entra credentials and authentication tokens. A hardware-bound signature that never leaves the enrolled device gives that kit no reusable login secret to capture. Already-issued tokens from a completed legacy sign-in still have to be revoked. With MFA 2.0 as the only method, that stolen session is only that login: once it expires it cannot be reused, and the attacker cannot add a device.

What could login MFA not undo after COLD RELIC already had Super Admin?

Login MFA could not undo domain-wide Customer Takeout, mass Drive sharing, roster export, admin log queries, or Alert Center deletion after COLD RELIC already held a Super Admin session. Those steps are post-authentication privilege abuse. The prevention win is blocking the push-fatigued Super Admin sign-in and the later Entra token phishing so those admin APIs are never reached under attacker control.

How should teams remediate if attacker-enrolled passkeys already exist?

Teams should revoke every attacker-enrolled passkey, MFA device, backup code, and OAuth grant, kill active sessions, and only then reset passwords on accounts touched in a COLD RELIC-style intrusion. GTIG and Mandiant note reauthentication with unrevoked attacker-enrolled passkeys after password and session resets. Incomplete revoke leaves the operators holding legitimate-looking factors even when the original password is gone.