top of page

The Spy Who Stole a Familiar Face: How Cold Relic Turned One Cloud Breach Into Another

hace 13 minutos
9 min de lectura

The first building was heavily protected.

Its employees carried identification cards. Its security team monitored access. Its administrators controlled the most sensitive rooms.

But the spy wasn’t interested in stealing only what was inside.

He wanted something that would open doors elsewhere.

Names.

Relationships.

Conversations.

Documents.

The identities of people whom other organizations already trusted.

When he finally escaped with the stolen information, the first building’s security team believed the immediate danger was over.

But the spy had already begun planning his next operation.

He studied the employees’ correspondence.

He examined their professional contacts.

He discovered which people regularly exchanged confidential information.

Then he selected a familiar identity.

And approached a second building.

The guard recognized the name.

The message seemed credible.

The request appeared to come from a trusted professional contact.

The spy had transformed one organization’s stolen identity into a weapon against another.

On November 18, 2026, Google’s Mandiant and Threat Intelligence Group (GTIG) detailed how the Russia-linked espionage group Cold Relic used information stolen from one U.S. organization to compromise a second.

The campaign illustrates a particularly dangerous combination: cloud identity compromise, persistent authentication access and social engineering built around genuine professional relationships.


Phase 1 — The Spy Chooses His First Building


The operation began in late 2025.

Cold Relic compromised the cloud productivity environment of an unnamed U.S.-based organization.

The group, previously tracked as COLDRIVER and UNC4057, is also known as Star Blizzard, Callisto and Gossamer Bear.

Its historical targeting has included government officials, defense contractors and nongovernmental organizations across Europe and North America.

Its objective has frequently involved obtaining information of potential strategic value to the Russian government.

In this operation, the attackers approached through a residential proxy network geographically close to the victim.

That infrastructure helped the activity appear more consistent with access originating near the legitimate user.

But the decisive weakness was not the network location.

It was the compromised identity of a highly privileged administrator.

The spy wasn’t searching for an unlocked window. He was looking for the person who held the master keys.


Phase 2 — The Guard Receives Too Many Requests


The attackers possessed stolen credentials for a Google Workspace Super Admin account.

But the account was protected by multifactor authentication.

The password alone was not enough.

The spy therefore began repeatedly requesting authorization.

In the real attack, Cold Relic used MFA push notification fatigue.

Repeated authentication prompts were sent until the victim accepted a request.

Once that happened, the attackers authenticated successfully.

The account’s Super Admin privileges provided access to highly sensitive administrative capabilities.

The incident demonstrates why push-based MFA, although substantially better than password-only authentication, can remain vulnerable to social engineering.

A legitimate user can mistakenly approve a malicious request.

The guard didn’t hand over the keys because the spy broke the lock. He handed them over because he eventually approved the wrong request.


Phase 3 — The Spy Makes His Own Keys


Once inside, the attackers did not rely exclusively on the initial stolen password.

They registered their own authentication device within the compromised environment.

Google also observed the use of passkeys and MFA backup codes as alternative authentication mechanisms.

This created additional ways to maintain access.

The distinction matters because incident response cannot always stop at changing a password.

If attacker-controlled authentication methods remain registered, the adversary may still possess a valid route into the account.

Google observed Cold Relic reauthenticating through attacker-enrolled passkeys that survived password and session resets.

In the metaphor, the spy had copied the master key.

Then he registered himself as an authorized keyholder.

Even after the original key was replaced, his own remained valid.

The intruder wasn’t merely hiding inside the building. He was making sure the building would continue recognizing him as an authorized visitor.


Phase 4 — The Entire Archive Disappears


With Super Admin access, Cold Relic began collecting organizational information.

Google reported that the attackers downloaded the contents of the organization’s cloud storage drive.

They used multiple methods, including sharing files in bulk with an attacker-controlled email account and downloading documents directly.

They also used Google Takeout to export employee lists and domain-wide data.

The stolen information included material that could support future targeting.

Employee directories.

Organizational records.

Documents.

Professional relationships.

Potentially sensitive operational information.

The significance was not limited to the confidentiality of the first victim’s files.

The information could help the attackers understand how the organization interacted with other people and companies.

The spy had stolen the archive. But what interested him most was the address book inside it.


Phase 5 — The Spy Studies the Security Cameras


Cold Relic also investigated the victim’s defensive visibility.

Google observed the attackers accessing the Admin Console and querying security investigation and email audit logs.

They viewed, modified and deleted Alert Center notifications.

These actions were intended to reduce evidence of the intrusion and understand what defenders might detect.

This is an important escalation beyond simple credential theft.

The adversary was not merely using a compromised account.

It was taking advantage of administrative privileges to interfere with the mechanisms designed to reveal unauthorized activity.

In the building, the spy entered the security office.

He studied the surveillance recordings.

He examined the incident reports.

And he removed warnings that might alert the guards.

The stolen master key opened more than the archive. It also opened the room where the guards kept their evidence.


Phase 6 — The Intruder Refuses to Leave


The first organization’s defenders eventually removed the attackers from the environment.

But Google observed continued Cold Relic activity over several months.

The group synchronized email through OAuth-connected external clients.

It reused attacker-enrolled authentication methods.

And it continued searching cloud documents of interest.

Even after the initial access was removed, the group impersonated employees of the compromised organization in renewed social engineering attempts.

This reveals two different forms of persistence.

The first is technical: retaining access through authentication devices, passkeys, tokens or connected applications.

The second is social: continuing to exploit the identities and information already stolen.

An organization may eliminate the first without eliminating the second.

The attacker can lose access to the building while retaining copies of its correspondence and employee directory.

The spy had been expelled. But he still knew everyone’s name, their relationships and how to imitate them.


Phase 7 — The Second Building Receives a Familiar Visitor


In 2026, Cold Relic targeted another unnamed U.S. organization.

This organization used Microsoft Entra ID as its federated single sign-on identity provider for Google Workspace.

The attackers impersonated an employee of the first compromised organization.

They contacted the new target through a secure messaging application.

Their message referred to a technical issue and offered a document containing relevant details.

The pretext appeared credible because the attackers could draw on authentic information stolen during the earlier compromise.

Google assessed that this knowledge of professional relationships increased the likelihood of successful phishing.

The spy had prepared carefully.

He knew the name he needed.

He knew how the person communicated.

He knew the context that would make the message believable.

When the second building received his request, it appeared to come from someone familiar.

The first organization’s stolen reputation became the spy’s passport into the second.

Phase 8 — The Document That Wasn’t a Document

The message included phishing links presented as a way to access information about a technical issue.

But the purpose was to capture authentication material.

The attackers sought Microsoft Entra ID credentials and authentication tokens.

This was particularly significant because Entra ID served as the identity provider for the organization’s Google Workspace environment.

Compromising the identity system could therefore provide access to cloud resources connected through federated authentication.

The attack combined human deception with identity infrastructure abuse.

The victim was not simply being asked to open a malicious attachment.

They were being directed toward an authentication flow that could expose their credentials and tokens.

In the building, the supposed colleague handed the guard a document.

The guard believed it contained information needed to resolve a technical problem.

But the document led him to a counterfeit identity checkpoint.

The spy didn’t need to force the second door open. He persuaded someone to reveal the credentials that opened it.


Phase 9 — Another Key Is Registered


After obtaining the necessary credentials, Cold Relic successfully authenticated to the second organization’s environment.

The attackers registered a new MFA device in the single sign-on system.

That device provided a means of retaining access to the compromised user account during the intrusion.

The pattern resembled the first compromise.

Obtain credentials.

Complete authentication.

Register attacker-controlled authentication methods.

Use the resulting access to collect information.

Google reported that Cold Relic performed bulk downloads of files accessible through the compromised user’s cloud drive.

The available evidence does not establish that the attackers gained unrestricted administrative control over the second organization.

The confirmed scope involved the compromised user identity and the cloud files that account could access.

The spy had entered the second building. And once again, he immediately arranged for the guards to recognize his own key.


Phase 10 — The Spy Changes His Face Again


After compromising the second organization, the attackers changed the profile name of the messaging account used to initiate contact.

Google suspected that the new persona could be used to approach additional organizations within the same professional network.

That suspected next step was not presented as a confirmed third compromise.

Nevertheless, it illustrates the potential for a cascading campaign.

The first breach provides documents and identities.

Those identities support a convincing approach to a second victim.

The second breach may provide additional contacts, correspondence and operational information.

The attacker can then repeat the process.

The attack becomes less about one isolated organization and more about a connected network of trust.

The spy didn’t need a new disguise from scratch. Every building he entered could provide the materials for the next one.


Phase 11 — Why Professional Trust Became the Weakest Door


The most distinctive feature of this campaign was not an exotic malware family or an unknown software vulnerability.

It was the weaponization of authentic relationships.

Cold Relic used real organizational information to make fraudulent communications appear legitimate.

A familiar employee name is more convincing than an unknown sender.

A message referencing a real technical concern is more persuasive than a generic phishing email.

A document-sharing request from an established professional contact can seem routine.

That is why compromising one organization’s cloud environment can create risks for its partners.

The original victim’s incident may become a source of intelligence for subsequent attacks.

This is especially important for organizations that regularly exchange sensitive information with contractors, suppliers, researchers, government agencies or professional associations.

The spy wasn’t attacking the second building’s walls. He was attacking the trust between the people who worked inside both buildings.


Phase 12 — How Defenders Can Stop the Impostor


Defending against this kind of campaign requires attention to both technical identity controls and human relationships.

1. Deploy phishing-resistant MFA. Prefer appropriately configured FIDO2 security keys or passkeys that resist credential phishing and adversary-in-the-middle attacks.

2. Monitor new authentication registrations. Alert on unexpected MFA devices, passkeys, backup methods and changes to authentication policies.

3. Review all persistence mechanisms after a compromise. Resetting passwords alone may be insufficient. Investigate registered devices, active sessions, tokens, OAuth grants and external email clients.

4. Protect privileged cloud accounts. Apply least privilege, separate administrative identities and use stronger controls for Super Admin accounts.

5. Detect abnormal cloud data movement. Investigate mass file sharing, unusual Google Takeout exports and large downloads from cloud drives.

6. Protect security telemetry. Monitor changes to alert configurations, audit visibility and administrative security settings. Preserve logs in systems attackers cannot easily alter.

7. Verify unusual requests independently. When a known contact unexpectedly shares a sensitive document or asks for authentication, confirm the request through a previously established channel.

8. Consider third-party exposure. If a partner suffers a breach, assess whether stolen correspondence or employee information could be used to impersonate its staff.

9. Strengthen federated identity security. Monitor Entra ID sign-ins, authentication changes and suspicious access to connected Google Workspace resources.

10. Prepare for post-compromise impersonation. Warn relevant employees and trusted partners when stolen identities or correspondence may be used in subsequent phishing attempts.

These controls address different parts of the same operation.

Some prevent the spy from acquiring the first key.

Others detect unauthorized access.

And others stop him from using a stolen identity to deceive the next organization.

The guards must protect not only their own doors, but also the trust that connects their building to everyone else’s.


The Final Scene — The Face Behind the Mask


The second building was quiet.

An employee received a message from someone they recognized.

The sender appeared to work at a familiar organization.

The conversation seemed ordinary.

The document sounded relevant.

Nothing about the request immediately suggested a stranger.

But behind the familiar name was a spy.

Months earlier, he had compromised another organization’s cloud environment.

He had stolen its employee records.

He had examined its communications.

He had copied its documents.

And he had learned which identities would be most useful.

The first organization’s defenders eventually removed his access.

But they could not undo everything he had already learned.

Now he stood inside the second building.

Another authentication device had been registered.

Another cloud drive had been accessed.

And the messaging account had already adopted a different identity.

The spy looked at the next name in his stolen directory.

Another contact.

Another relationship.

Another door.

Google’s investigation demonstrated how one cloud compromise could become the foundation for another, not merely through stolen technical credentials, but through the exploitation of genuine professional trust.

The lesson reaches beyond Cold Relic.

Organizations do not operate in isolation.

Their employees communicate.

Their systems exchange information.

Their partners trust familiar identities.

And when an attacker steals enough of that context, the next intrusion may begin with a message that looks entirely ordinary.


The first breach gave the spy the keys.

But the stolen identities taught him how to knock on the next door without raising suspicion.

Because the most dangerous disguise isn’t the one that hides an intruder’s face. It’s the one that makes everyone believe they’ve seen that face before.


 
 
 

Comentarios


bottom of page