Cold Relic is a Russian state-sponsored espionage group, formerly tracked as COLDRIVER and UNC4057 and also known as Star Blizzard, Callisto and Gossamer Bear. It used data stolen from one US organization's Google Workspace tenant to break into a second US organization. Google Threat Intelligence Group (GTIG) and Mandiant disclosed the two linked intrusions on 17 November 2026. Mandiant responded to both. The first happened in late 2025 and the second in 2026. Google says the attackers "directly weaponized data, the corporate staff directory, and identities" taken in the first breach to go after the second. Google cites the US Department of Justice's assessment that the group operates within Russia's Federal Security Service (FSB). Neither victim has been named. Google has given no victim counts or data volumes, and neither organization has made a public statement or filed with a regulator.
What Happened
Google's account describes a chain of linked intrusions. A single cloud compromise became, in Google's words, "a springboard for ongoing third party compromise and espionage."
The first intrusion (late 2025). The attackers got into the first organization's cloud productivity environment with stolen credentials for a Super Admin account. They also bombarded the account holder with MFA push requests until one was accepted. Their traffic ran through a US residential proxy "in close geographical proximity to the victim," which made the logins look local. Once inside, they enrolled a device the tenant had not seen before and added their own passkeys, MFA backup codes and authenticator devices. The Circuitry reports that this access lasted several months.
The second intrusion (2026). The group used names, identities and relationship details from the first victim to impersonate an employee and approach the second organization. The Circuitry reports that a phishing link's parameters showed the operators likely drew on relationship details from the earlier victim. Accounts differ on the channel. The Circuitry describes a work chat. Czech outlet Kyberkit describes an encrypted messenger and links that steal Microsoft Entra ID tokens. Analysis platform Mallory, summarizing Google, reports that after one compromise an attacker-controlled encrypted messaging account changed its profile name to a different person.
Google says it has tracked and disrupted Cold Relic phishing since 2015. It has also seen new phishing operations from the group as recently as November 2026.
What Was Taken
Google has not published record counts or data volumes for either victim. It is clear about the kinds of data taken from the first organization:
- The full staff directory, which was exported and later used to pick and impersonate contacts.
- Cloud storage files. With Super Admin rights, the operators shared files out to an account they controlled and downloaded others directly.
- Email and chat content, which Google describes as standard collection targets for the group.
- A possible copy of the whole tenant. The Circuitry reports that the operators started a tenant-wide export through Google Workspace Customer Takeout. Kyberkit says data left through that export, among other routes.
The operators also read the security and mail audit logs, then deleted the Alert Center notices that recorded the break-in. Google has not said in detail what was taken from the second organization.
Why It Matters
Cold Relic is known for patient, well-researched phishing against government officials, defense contractors and NGOs in Europe and North America. This case shows what the group does after it gets in. Each breach supplies real email threads, real staff names and real working relationships, and those make the next lure much more convincing. A breach therefore puts every partner, supplier and contact of the victim at risk, not only the victim.
Persistence is the second problem. The Circuitry reports that after the first organization forced password resets, unrevoked sessions and passkeys enrolled by the attackers still worked. Kyberkit says these methods survived both password changes and sign-outs. Mallory's summary of Google's guidance warns that "password resets alone may leave the actor's access intact." An incident response plan that stops at resetting credentials would have left this group inside.
Kyberkit notes that Google has not published indicators of compromise. Google has instead offered detection tips for Workspace admin logs, mapped to MITRE ATT&CK. Defenders will need to hunt for the behavior described here rather than match known indicators.
The Attack Technique
- Initial access: stolen Super Admin credentials and MFA push fatigue in the first case. Impersonating a known contact, plus phishing links, in the second.
- Evading detection: residential proxies close to the victim, so login location alerts do not fire.
- Persistence: rogue MFA devices, passkeys and backup codes, and hijacked sessions that stayed valid after password resets.
- Collection and exfiltration: bulk Drive downloads, sharing files to accounts outside the organization, a staff directory export, and a Customer Takeout export of the tenant.
- Covering tracks: reviewing audit logs and deleting Alert Center notices.
- Expansion: stolen relationships and pretexts reused against the next target. Mallory adds that the group also used credentials stored in the cloud to widen its access.
What Organizations Should Do
- Give admins phishing-resistant MFA. Require FIDO2 security keys for Super Admin and other privileged accounts, and turn off push approval for them.
- Control new authenticators. Alert on, and where possible require approval for, new MFA devices, passkeys and backup code generation, especially on admin accounts.
- Revoke everything during incident response. After a compromise, revoke all sessions, OAuth tokens and app grants, and remove any passkeys, MFA devices and backup codes you don't recognize. A password reset is not enough.
- Watch for data leaving in bulk. Alert on Customer Takeout exports, directory exports, large-scale sharing of Drive files to outside accounts, and deletion of Alert Center items. Apply cloud data loss prevention policies.
- Don't trust login location alone. Residential proxies near the victim defeat location checks. Combine location with signals such as unfamiliar devices, unusual ASNs and new authenticator enrollments.
- Warn your partners after a breach. If your tenant is compromised, tell close partners and contacts that messages appearing to come from your staff may be impersonation. Tell your own staff to confirm unexpected requests from known contacts through a separate channel.
Sources: Russian COLD RELIC Hackers Use Stolen Google Workspace Data to Brea... | COLD RELIC: A Case Study in Impersonation and Compromise Google Cl... | Russian group COLD RELIC uses stolen contacts to break into new US... | Google links COLD RELIC's 2025 US breach to a second intrusion — Th... | COLD RELIC Uses Professional Impersonation and Cloud Identity Abuse... | Cold Relic kompromittiert Google-Workspace über Super-Admin und Ket... | COLD RELIC (COLDRIVER): únava z MFA a podvržené passkeys | GTIG: COLD RELIC Used One US Victim to Hit a Second TSN