Home/Blog/Cold Relic turned one Google Workspace breach into two
blog

Cold Relic turned one Google Workspace breach into two

November 18, 2026 · Security News

Cold Relic Google Workspace breach: Super Admin takeover used to phish a second company via Entra ID

Russia-linked espionage group Cold Relic hijacked a Super Admin account in a U.S. organization's Google Workspace in late 2025, stole the contents of its entire cloud drive, then used the staff directory to breach a second U.S. firm through Microsoft Entra ID in 2026, Google's Mandiant reports. No software flaw was named, so there is nothing to patch. Audit admin passkeys, MFA devices and OAuth mail grants this week: they kept the attacker inside after passwords were reset.

Key takeaways

  • Cold Relic (COLDRIVER, Star Blizzard) stole the contents of a U.S. organization's entire Google Workspace drive in late 2025 via a Super Admin account taken over with a push-notification fatigue attack.
  • A registered device, passkeys, backup codes and OAuth mail sync kept the actor inside for months after the theft.
  • The actor deleted Alert Center notices and used Google Takeout to export the staff roster, which fed the next phishing wave.
  • In 2026 a messaging-app lure harvested Entra ID credentials and tokens at a second U.S. firm; the attacker added its own MFA device in the SSO layer and bulk-downloaded a user's drive.
  • No CVE, malware or IOCs were disclosed: defense means auditing admin roles, authentication methods and OAuth grants on both Workspace and Entra ID.

What happened in the Cold Relic Google Workspace intrusions?

Two U.S. organizations lost data to the same Russian espionage crew, and the second break-in was built on what was stolen in the first. Google’s Mandiant incident responders and the Google Threat Intelligence Group (GTIG) investigated a 2026 intrusion at a U.S.-based organization and traced it back to a different U.S. organization that Cold Relic compromised in late 2025.

The corporate staff directory and employee identities stolen from the first victim became the raw material for the attack on the second.

Google tracks the actor as Cold Relic. Defenders know it under other names: COLDRIVER (which Google lists as the former label), UNC4057, Callisto, Gossamer Bear and Star Blizzard. Same operators, different vendor naming schemes.

The detail that matters most for anyone running a tenant: there is no CVE. Google’s findings, as reported by Ravie Lakshmanan at The Hacker News, name no vulnerability, no malware family, no file hashes and no network indicators. Every step used features Google and Microsoft ship on purpose: admin consoles, data export, passkeys, OAuth mail sync and federated sign-on.

Nothing to patch, no signature to deploy. The fix lives in how you configure and watch identity.

How did Cold Relic get in and stay in after password resets?

Cold Relic got in with a stolen Super Admin password and one employee who tapped “approve,” then layered several independent ways back in so no single reset could lock it out. Google does not say how the credentials were stolen in the first place.

The first login came through a residential proxy network located close to the victim geographically. Residential proxies route traffic through IP addresses assigned to ordinary home broadband connections. To a sign-in log, the session looks like a staff member working from home in the same metro area. Country blocks, IP reputation feeds and impossible-travel rules all see a plausible local sign-in and fire nothing.

MFA was push-based. The actor sent prompts until the account holder accepted one, which Google calls a push-notification fatigue attack. It works because the prompt asks a tired human a yes-or-no question with no context. One wrong tap on a Super Admin account handed over the most powerful role in the tenant.

Then came the persistence work, which is the real lesson:

  • A registered device. The actor enrolled its own device inside the tenant.
  • Passkeys. Google says the attacker’s passkeys were never revoked, survived password and session resets, bypassed device registration controls, and were used to sign back in.
  • MFA backup codes. Held as another spare key.
  • OAuth mailbox sync. Inboxes were synced to external mail clients.

Each of these survives the standard “reset the password, kill the sessions” response. A passkey is a separate credential bound to the account, so a new password does not touch it.

OAuth mail clients typically hold refresh tokens and keep pulling mail without producing an interactive login, which is why a sign-in review alone misses them. Google says the actor kept access for several months after the bulk theft.

What data left, and how was the evidence cleaned up?

Cold Relic exfiltrated the entire organization’s cloud storage, the full employee roster and domain-wide data, then used the Admin Console to see what defenders could see and erase what they would have seen first.

Google describes the drive theft as two-pronged: “Cold Relic used the privileges of the compromised Super Admin account to download the contents of the entire organization’s cloud storage drive by both mass sharing files to a Cold Relic-operated email account and downloading files directly.”

Mass sharing to an outside address and direct downloads leave different audit trails. A team watching only one sees half the picture.

The roster and domain data came out through Google Takeout, the legitimate export tool most people know for downloading their own photos. With Super Admin rights, it exports far more. No malware needed, just a button that exists for compliance and migration.

Later collection narrowed. Google says the actor went after documents relating to credentials, email archives and personnel information. Credential documents feed further access. Personnel files and email archives feed the next pretext.

The cleanup shows insider-level fluency with the tenant’s own admin tooling. Per Google, Cold Relic “queried Admin security investigation and email audit logs to check defensive visibility, and viewed, modified, and deleted Alert Center notifications to remove forensic evidence.” It checked whether anyone was watching, then deleted the alerts that would have flagged the first break-in.

Deleted alerts are not the end of the evidence. Alert Center is a notification console, not the tenant’s only record. Admin, login, Drive, OAuth token and Takeout events sit in separate audit logs, and the attacker’s own log queries are admin actions too.

Responders who export everything early, before retention windows roll over, usually still have a case. That is why a whole-drive export plus deleted alerts calls for a proper data breach response, not just a login reset.

How did one breach become the phishing for the next company?

The first breach became the second one’s address book. Google puts it plainly: “Once Cold Relic compromises a target, it uses stolen emails and data to establish authentic pretexts and impersonate contacts when reaching out to subsequent targets.”

After the first organization evicted the actor, Cold Relic did not move on. It used email accounts impersonating that organization’s employees and aimed them at the victim’s own staff and at associates in the same professional circles.

A stolen directory tells you who reports to whom, who works with which partners, and how people sign their mail. That makes the fake far harder to spot than a generic phish.

The second victim ran Google Workspace with Microsoft Entra ID as its federated single sign-on identity provider. The approach came over a secure messaging app (Google does not name it), from an account posing as an employee of the first victim.

The lure was a shared document supposedly about a technical issue. The link led to a page built to harvest Entra ID credentials and authentication tokens.

Tokens are the prize. An adversary-in-the-middle phishing page can proxy the real Microsoft sign-in, let the user complete MFA, and capture the resulting session. Replaying that session avoids a fresh MFA challenge without breaking any cryptography.

Google does not name the kit used here. We covered other identity tricks that skip the password in our piece on Microsoft device code phishing kits.

Once in, the actor repeated its habit. “Upon successful authentication using the stolen credentials, Cold Relic registered and used a new MFA device in the SSO solution to retain access to the user account throughout the intrusion,” Google wrote.

The device lived in the identity provider, not in Google. Because Workspace trusted Entra ID for sign-in, owning the Entra account meant owning the route into Workspace. The actor then bulk-downloaded the files available in that user’s cloud drive.

The blast radius was smaller this time: one federated user rather than a Super Admin, so one user’s accessible drive rather than the whole tenant.

The messaging-app account was later renamed to a different persona. Google says the new identity was likely used to target other organizations in the same professional network.

Who is Cold Relic, and should a smaller business care?

Cold Relic is a Russian state espionage group whose usual targets are government officials, defense contractors and NGOs across Europe and North America. Under the Star Blizzard name, the UK’s NCSC and allied agencies have publicly attributed it to Russia’s Federal Security Service (FSB).

A 40-person firm might read that target list and relax. Bad idea. The way into a ministry or a defense prime often runs through the consultancy, law firm, think tank or supplier that emails it every week.

Google’s own framing: “By obtaining access to strategic personnel rosters and operational documents, Russian intelligence services turn legitimate business communications into attack vectors against trusted partners.” If your staff appear in a partner’s directory, you are in the chain whether you chose to be or not.

The methods are also cheap. Push-notification fatigue, residential proxies, passkey enrollment and Takeout work against any tenant with a sloppy admin setup. A criminal crew can copy this playbook tomorrow.

What should an IT manager check this week?

Start with authentication methods and admin roles, not passwords. This list maps to what Cold Relic actually did. Google published no indicators for this case, so rely on configuration and logs rather than blocklists.

Google Workspace

  • Count your Super Admins. Most SMBs need two, one of them a break-glass account. Move daily admin work to narrower delegated roles.
  • End push-only MFA for admins. Require security keys or passkeys enrolled by your own team on privileged accounts, and use number matching wherever push remains.
  • Review every admin’s passkeys, security keys, registered devices and backup codes. Remove anything nobody can explain.
  • Audit OAuth grants. Look for third-party mail clients with mailbox access on admin and executive accounts, and restrict which apps may request Gmail scopes.
  • Search the audit logs for Takeout exports, bulk sharing to a single outside address, unusual download volumes, admin queries against security investigation and email audit logs, and edited or deleted Alert Center notices.
  • Restrict Takeout for accounts with no business need for it.

Entra ID federated to Workspace

  • Check new MFA registrations at the identity provider. In the second intrusion, the attacker’s device lived in Entra ID. A Google-side review would miss it.
  • Check log retention. Find out how long your Entra ID tier keeps sign-in and audit logs; on most tiers it is weeks, not months. Cold Relic stayed for months at the first victim. Forward logs to storage or a SIEM now.
  • Bind sessions to managed devices with Conditional Access and phishing-resistant MFA, which blunts replayed tokens.

People and partners

  • Treat an unexpected shared document “about a technical issue” from a known contact, especially on a messaging app, as a possible stolen-trust lure. Verify through a channel you already have.
  • If a partner reports a breach, assume its directory included your people and brief those staff by name.
  • An MFA prompt nobody triggered is an incident to report, not a nuisance to dismiss.

If you suspect compromise, eviction is an inventory-and-revoke job. Preserve the current list of sessions, passkeys, security keys, MFA devices, backup codes, OAuth grants and mail delegation first. Then remove the attacker’s items one by one, on both sides of a federated trust.

Our account takeover recovery work follows that order, because a reset done first wipes out the evidence of what to revoke.

Practitioner’s note: When we test Workspace and Entra tenants, the repeat finding is not a missing patch. It is too many Super Admins, at least one still on push approval, and admin passkey lists nobody has ever opened. OAuth mail clients granted years ago by departed staff come a close second.

Look at authentication method changes before bulk downloads: an unexplained passkey or MFA enrollment is usually the moment a stolen session became a foothold. And because the second breach started with a believable message under a real colleague’s name, a social engineering assessment that impersonates your actual partners shows which staff would open that shared document.

What is still unknown?

Google has left several gaps in the public account. It has not named either organization or its sector. It has not said how the first Super Admin password was stolen, which residential proxy provider was used, which messaging app carried the lure, which phishing kit harvested the Entra ID tokens, or how the first victim finally detected and evicted the actor.

The renamed persona suggests other organizations in the same professional network may already have been approached. None has been identified publicly.

What is clear: Cold Relic needed no exploit. It needed one approved push, one overpowered account and a response that reset passwords without listing everything else attached to the account.

FAQ

Is Cold Relic the same group as Star Blizzard and COLDRIVER?

Yes. Google tracks the group as Cold Relic and lists COLDRIVER as its former name. Other vendors call it UNC4057, Callisto, Gossamer Bear and Star Blizzard.

Did this attack use a Google or Microsoft vulnerability?

No vulnerability was disclosed. The actor used stolen credentials, push-notification fatigue, phishing and legitimate features such as Google Takeout, passkeys and OAuth mail sync.

Can a passkey survive a Google password reset?

Yes. A passkey is a separate credential, so resetting the password and revoking sessions does not remove it. It has to be deleted from the account explicitly.

Does Entra ID SSO protect Google Workspace from phishing?

Not by itself. Federation moves the front door to Entra ID, so an attacker who phishes Entra credentials and tokens and registers an MFA device there gets into Workspace too. Check MFA registrations at the identity provider.

What should I check after a suspected Workspace admin takeover?

List and preserve every session, passkey, security key, MFA device, backup code and OAuth grant before revoking anything. Then search audit logs for Takeout exports, mass external sharing and deleted Alert Center notices.

Not sure who still holds a passkey or OAuth grant on your admin accounts? Our account takeover recovery team can inventory, revoke and confirm the attacker is really gone.


← All articles

get in touch

Tell us what you're running

Scoping is free. We reply within one business day, and under 30 minutes for active incidents.