Blog / Front Line — part 23 of 25

COLD RELIC: MFA fatigue, rogue passkeys and a second breach

· urgentic · 5 min read

Short read

This post covers COLD RELIC, the Russia-linked espionage group that Google's Mandiant and Threat Intelligence Group say turned one cloud breach into the starting point for another. The group got into a Super Admin account with stolen credentials and an MFA fatigue attack, enrolled its own passkeys and devices, and later used stolen staff details to phish a second organisation.

You'll learn how the attacks worked, why resetting a password wasn't enough to get the group out, and the identity checks Google recommends for your own cloud tenant.

An MFA fatigue attack on a single admin account was all it took. On 17 November, Google's Mandiant and Threat Intelligence Group (GTIG) published a case study on COLD RELIC, the Russian cyber espionage group previously tracked as COLDRIVER and UNC4057, and also known as Star Blizzard, Callisto and Gossamer Bear. It shows how the group took a cloud compromise at one US organisation in late 2025 and used what it stole to break into a second US organisation in 2026.

There's no exotic malware or zero-day here. Every step abused normal cloud identity features: sign-ins, MFA prompts, passkeys, OAuth grants and file sharing. That's why it matters to anyone running Google Workspace or Microsoft 365.

How the first breach worked

According to Google, COLD RELIC signed in through a residential proxy network in the US, close to the victim's location, so the logins wouldn't look out of place. It used stolen credentials for a Super Admin account and bombarded the user with MFA push notifications until one was accepted.

Once inside, the group:

  • registered a device model never seen before in the tenant, and used passkeys, MFA backup codes and rogue MFA devices to keep its access
  • downloaded the contents of the organisation's entire cloud storage drive, both by mass-sharing files to an email account it controlled and by downloading them directly
  • queried admin security and email audit logs to see what defenders could see, and viewed, changed and deleted Alert Center notifications to remove evidence
  • exported full employee lists and domain-wide data through Google Takeout
  • synced mailboxes to external email clients via OAuth

The Hacker News reports that the attackers kept this access for several months.

Why a password reset didn't get them out

One detail stands out. GTIG says that in at least one case the victim spotted the intrusion and reset passwords for the affected accounts. But because the attackers had hijacked sessions and the organisation didn't also revoke them, COLD RELIC kept its access. Passkeys the attackers had enrolled also survived password and session resets.

The lesson is simple: when an account is compromised, changing the password is only the start. Sessions, refresh tokens, OAuth app grants, and any unknown passkeys, backup codes or MFA devices all have to go too.

From one victim to the next

The second intrusion shows why this is about more than one organisation. COLD RELIC contacted an employee at a second US organisation over an encrypted messaging app, posing as someone from the first victim. The conversation was about a technical issue, and the group shared links to a "document" that actually harvested Microsoft Entra ID credentials and authentication tokens.

That organisation used Entra ID as its single sign-on provider for Google Workspace. After signing in with the stolen credentials, the group registered a new MFA device in the SSO system and bulk-downloaded files from the user's cloud drive. Google says the messaging account was then renamed to a new persona, probably for the next target.

Google also notes that the damage was made worse because users had stored passwords and other secrets in cloud storage, which let the attackers move on to further accounts, including personal ones.

In other words, a breach at a partner, supplier or client can hand an attacker exactly the names, relationships and context needed to fool your staff.

What to check in your own tenant

Google's report includes a clear set of recommendations. The main ones are:

  • Move to phishing-resistant MFA. Replace SMS, voice and push approvals with FIDO2 or WebAuthn security keys, especially for admins. These can't be approved through fatigue or intercepted by reverse proxies.
  • Control MFA and passkey enrolment. Require an existing admin or out-of-band check before a new MFA device or passkey is added, and review enrolments on admin accounts.
  • Revoke, don't just reset. Build session, token and OAuth revocation into your account compromise playbook, across both Google Workspace and Entra ID, and remove unauthorised passkeys, backup codes and MFA devices.
  • Audit third-party OAuth grants. Flag unexpected desktop email clients or external mail services with access to mailboxes.
  • Restrict admin logons from residential proxies, consumer ISPs and VPNs using conditional access.
  • Use data loss prevention to warn on or block mass external sharing and files that contain passwords or staff rosters.

The detections that would have caught it

Google also lists the log events that give this activity away. Most organisations already collect these logs; the question is whether anyone is watching them:

  • many MFA challenges for one user in a short time, followed by a successful sign-in
  • new passkey enrolments or device registrations, especially from unexpected locations or straight after an MFA fatigue pattern
  • sign-ins using MFA backup codes
  • new OAuth authorisations granting broad mail access to unapproved clients
  • any tenant-wide export, such as a Google Takeout job
  • spikes in file downloads or external sharing by a single user, particularly to free email providers
  • admins viewing or deleting security alerts

Each of these on its own may look like a normal admin task. Together, and on a privileged account, they tell a clear story. That's why identity logs need the same round-the-clock attention as endpoint and network data, with someone able to connect the events quickly. It's the approach behind the urgentic platform.

The bottom line

COLD RELIC didn't need to break any software. It needed one tired user to tap "approve", a tenant that let new authenticators be added without question, and logs that nobody was reading in time. Phishing-resistant MFA closes the front door; watching identity events, and revoking everything when something goes wrong, stops an intruder from staying for months.

  • COLD RELIC
  • COLDRIVER
  • MFA fatigue
  • passkeys
  • Google Workspace
  • Entra ID
  • identity security
  • cloud security
  • SOC