Russia's COLD RELIC Turned One Breached Google Workspace Into the Lure for the Next Company. What a Workspace Admin Checks in the Audit Logs Tonight
If you administer Google Workspace, do five things tonight. Open the Admin console, go to Reporting, then Audit and investigation, and look at the last 90 days for these events. One: any Customer Takeout created in Admin log events (CUSTOMER_TAKEOUT_CREATED). A tenant-wide export needs a Super Admin, and you should be able to name who asked for it. Two: any User List Download in Admin log events. Three: any Batch Delete Operation Performed on Alert Center alerts, which is someone deleting the warnings meant for you. Four: in User log events, passkey enrollments, new device registrations, and logins whose challenge type was Backup code. Five: in OAuth log events, any Authorize event that grants the full mail scope (https://mail.google.com/) to a desktop mail client or outside mail service you did not approve. Then, if you ever have to throw someone out, revoke sessions, OAuth grants, passkeys, backup codes and MFA devices. A password reset alone did not evict this group.
Google's Threat Intelligence Group (GTIG) and Mandiant published the report on Tuesday, November 17, 2026: "COLD RELIC: A Case Study in Impersonation and Compromise," by Josh Murchie, John Wolfram, Joseph Spero, Peter Ukhanov, Fernando Tomlinson, Jr. and Wesley Shields. The Hacker News covered it on Wednesday, November 18. Everything below comes from Google's report unless we say otherwise.
Who COLD RELIC is
Google: COLD RELIC is "formerly tracked as COLDRIVER and UNC4057; also known as Star Blizzard, Callisto, and Gossamer Bear." It is "a Russian state-sponsored threat group that the US Department of Justice (DOJ) has assessed operates within Russia's Federal Security Service (FSB)." GTIG says it has tracked and disrupted the group's phishing since 2015, and that it "has observed evidence of new phishing operations as recently as November 2026."
The group's usual targets, in Google's words, are "government officials, defense contractors, and non-governmental organizations (NGOs) across Europe and North America." That is the group's history. It is not a description of the two victims in this report.
Who was hit
Two organizations, both described only as "US-based." Google does not give their sector or size, and we do not know them.
The first, compromised in late 2025. COLD RELIC took over its "cloud productivity environment," which the details (Customer Takeout, Alert Center, Super Admin) show was Google Workspace.
The second, compromised in 2026. It used Microsoft Entra ID as its single sign-on identity provider for Google Workspace.
How one breach became the next
Organization one: stolen Super Admin, push fatigue, then everything. COLD RELIC came in through "a residential proxy network in the US located in close geographical proximity to the victim," so the login looked local. It signed in with "the stolen credentials of a Super Admin account" and an MFA push-fatigue attack. Once the user approved a prompt, it "registered a device model previously unseen within the tenant" and kept access with "passkeys, MFA backup codes, and rogue MFA devices."
Then it took the tenant. It downloaded "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." It opened the Admin console and "queried Admin security investigation and email audit logs to check defensive visibility," and "viewed, modified, and deleted Alert Center notifications to remove forensic evidence of the initial intrusion." It "exported full employee lists and initiated a domain-wide data export via Google Workspace Customer Takeout."
It stayed for months. Google lists inbox sync "via OAuth to external email clients," reauthentication with "unrevoked attacker-enrolled passkeys that persisted across password and session resets and bypassed device registration controls," and targeted reads of documents about credentials, email archives and personnel. After it was removed, it used "email accounts impersonating employees of the victim organization" to go after internal staff and people in the same professional circles.
Organization two: a familiar name on a messaging app. COLD RELIC "impersonated an employee of the organization impacted in Case Study 1" and opened a conversation "over a secure messaging app." Google does not name the app. During the chat it sent links "designed to steal credentials and authentication tokens for Entra ID," which then showed the victim real-looking documents "relevant to the ongoing conversation in order to avoid raising suspicion." With the stolen login it "registered and used a new MFA device in the SSO solution" and did a "bulk download of files accessible in the compromised user's cloud drive."
So the stolen staff directory and email from organization one supplied the name, the relationship and the topic for organization two. Google also correlated a COLD RELIC redirector URL, submitted to VirusTotal in June 2026, "whose query parameter referenced the first victim organization." The group's links usually start at a compromised website that redirects to the real phishing page and carry a parameter such as ?file= or ?ref= naming a past victim or the lure.
Afterward, the attacker's messaging account "changed their profile name to a completely new person." GTIG believes that persona was "likely used to continue to target other organizations in the same professional circles."
What it looks for once inside. Credentials, personnel information, .pst email archives, and the raw material for the next lure: "organizational rosters, contact lists, resumes, and event schedules."
What is not known
The victims' sectors. Google says "US-based" and nothing more.
How the Super Admin password was stolen in the first case. Google says "stolen credentials" and does not say how they were obtained.
Which messaging app was used in the second case.
Any indicators. Google's report publishes no IP addresses, domains, URLs or file hashes, and neither does The Hacker News article. The redirector URL it mentions is described, not listed. So we have nothing to add to our feed for this one, and this post has no blocklist. The two network numbers Google names, AS5650 and AS20115, are given as examples of residential provider networks to restrict for admin logins. They are large US consumer ISPs, not attacker infrastructure. Do not block them for everyone.
What to check tonight, by name
All of these live under Admin console, Reporting, Audit and investigation (or Security, Security center, Investigation tool, if you have it). The event names in quotes are the ones Google's report uses.
Admin log events: "Customer Takeout Created" (CUSTOMER_TAKEOUT_CREATED). Any hit you cannot tie to a ticket is an incident. Google: this "can result in data exfiltration of an entire tenant." While you are there, decide whether end users need Google Takeout at all; you can turn it off per organizational unit (Google's help article "Allow or block Google Takeout").
Admin log events: "User List Download." Someone exported your directory. That is the lure list for your partners.
Admin log events: "Investigation Query." Who has been searching your audit logs? An admin you cannot name doing it is the attacker checking what you can see.
Admin log events: "Batch Delete Operation Performed" on Alert Center alerts. Google says to watch especially for a user deleting alerts about their own account, or deleting "Government-backed attack" alerts.
User log events: "Passkey enrolled" and "Device Registered," above all on admin accounts and right after a burst of MFA prompts. Then filter Login Challenge Type for "Backup code." Logins with backup codes from a consumer ISP or a proxy network are the pattern in this report.
User log events: MFA fatigue. Several failed login or MFA challenges for one user in a short window, followed by a success from the same IP.
OAuth log events: "Authorize" with the scope https://mail.google.com/ granted to a desktop mail client or outside mail service. Google names Thunderbird and Proton as examples; neither is bad in itself, the question is whether you approved it for that user.
Drive log events: spikes of "Download" by one user, external sharing spikes ("Visibility changed" to shared externally, especially to @gmail.com), and "Item search performed" for terms like .pst, .bak, password or cred.
Two more we would add, which Google's report does not mention for this group: look at the "Email forwarding address" attribute in User log events for forwarding to outside addresses, and check whether any account has app passwords, which bypass 2-step verification for the apps that use them. Neither is in Google's write-up of COLD RELIC; they are the same persistence idea by another door.
If you federate Workspace sign-in through Entra ID like organization two, run the same check on the Entra side: new authentication methods registered on a user, and sign-ins from residential addresses near the user.
What to change
From Google's list, in the order we would do them:
Phishing-resistant MFA for admins first. Google recommends hardware security keys over SMS, voice and push. Push is what organization one approved.
Require approval for new MFA methods and passkeys, by an existing admin or out-of-band check.
Remediation means revoke everything. Google: one victim "attempted to remediate COLD RELIC's access to their cloud tenant via password resets," but because sessions were not revoked, "COLD RELIC maintained access." Revoke browser sessions, refresh tokens and third-party OAuth grants in both Google Workspace and Entra ID, and remove unauthorized passkeys, backup codes and MFA devices.
Restrict admin logins from consumer ISPs, VPNs and residential proxy networks. COLD RELIC picks proxies near the victim so impossible-travel alerts stay quiet.
Stop keeping passwords in Drive. Google says secrets stored in cloud storage let the group reach more accounts, including users' personal ones. Data loss prevention rules can warn on files that look like credentials or rosters.
Tell your partners. If you were breached, your staff names are now someone else's phishing pretext. A message from one of your people on a messaging app, carrying a document link, is what organization two received.
How we know
We read Google's report in full and quote it above; The Hacker News article pointed us to it. We checked the Admin console navigation against Google's help page on user log events. Google publishes no indicators, so there was nothing to look up in our corpus or ingest into our feed, and we make no claim about seeing this activity. These are identity and cloud-console attacks; our edge sensors watch inbound web traffic to our own sites, so they would not see them, and we did not look. A search of our own blog index for COLDRIVER, Star Blizzard, Callisto and Gossamer Bear found no earlier post of ours on this group.
Read next: A Google Calendar Invite From Someone You Know Is Stealing Live Google Sessions Through Your MFA, the other recent case where a password reset did not evict the attacker.
Sources
Google Threat Intelligence Group and Mandiant (Josh Murchie, John Wolfram, Joseph Spero, Peter Ukhanov, Fernando Tomlinson, Jr., Wesley Shields), "COLD RELIC: A Case Study in Impersonation and Compromise," Google Cloud Blog, November 17, 2026: https://cloud.google.com/blog/topics/threat-intelligence/cold-relic-a-case-study-in-impersonation-and-compromise
The Hacker News, "Russian COLD RELIC Hackers Use Stolen Google Workspace Data to Breach Second U.S. Organization," November 18, 2026: https://thehackernews.com/2026/11/russian-cold-relic-hackers-use-stolen.html
Google Workspace Admin Help, "User log events": https://knowledge.workspace.google.com/admin/reports/user-log-events
Google Workspace Admin Help, "Allow or block Google Takeout": https://knowledge.workspace.google.com/admin/users/advanced/allow-or-block-google-takeout
Every indicator in this post is in the feed. Free.
1.58M+ IOCs, STIX 2.1 / TAXII, 88% novel vs ThreatFox, exploited-CVE leads ahead of CISA. No credit card — a free API key in 30 seconds, and you can audit every claim above against the live endpoints.



Comments