COLD RELIC Reused Stolen Workspace Data: What to Do
TL;DR · the short version
On 17 November 2026 Google Threat Intelligence Group and Mandiant published a case study of COLD RELIC, the group Google formerly tracked as COLDRIVER (also Star Blizzard, Callisto). Google says that in late 2025 the group got into a US organisation's Google Workspace with stolen Super Admin credentials and an MFA push-fatigue attack, stayed for months with its own passkeys and backup codes, exported employee lists and mail, and deleted alerts. In 2026 it impersonated one of that organisation's employees on an encrypted messaging app to phish someone at a second US organisation, stole Entra ID credentials and tokens and downloaded their files. Neither victim is named. Google calls the group Russian state-sponsored and cites the US DOJ's assessment that it operates within the FSB. The defence: phishing-resistant MFA, controls on passkey and MFA enrolment, and revoking sessions and OAuth grants on every reset.
On 17 November 2026, Google Threat Intelligence Group (GTIG) and Mandiant published a case study of COLD RELIC, the group Google formerly tracked as COLDRIVER and also known as Star Blizzard and Callisto. Google says the group broke into one US organisation's Google Workspace in late 2025, took its employee lists, mail and contacts, and in 2026 used that material to impersonate an employee and compromise a second US organisation. Google does not name either victim. It calls COLD RELIC a Russian state-sponsored group and cites the US Department of Justice's assessment that it operates within Russia's FSB.
Who this is for: security leaders and identity teams running Google Workspace or Microsoft Entra ID, especially in government, defence, policy, research and NGOs, and anyone whose staff talk to those organisations.
Based on Google's post (read in full on 19 November 2026) and The Hacker News' coverage of 18 November. PlayCISO is not affiliated with Google. Phishing domains, IP addresses and network ranges from the report are not repeated here; Google's post has what defenders need.
Confirmed vs not yet known (19 Nov 2026)
- Reported by Google (GTIG and Mandiant investigated the 2026 compromise): two US-based victims; the late-2025 Workspace intrusion; the 2026 Entra ID compromise built on data from the first; the techniques and detections below.
- Attribution, as Google words it: a "Russia-nexus cyber espionage group" and "a Russian state-sponsored threat group". The FSB link is the US DOJ's assessment, which Google cites; Google does not give its own confidence level for it.
- Not disclosed: who the victims are, their sectors, what data mattered most to the actor, how the Super Admin credentials were first stolen.
- Still active: Google says it saw "evidence of new phishing operations as recently as November 2026".
- Not a platform breach: nothing in the report describes a flaw in Google Workspace or Entra ID. It describes stolen credentials and legitimate admin features used against their owners.
How one breach fed the next
Victim 1: a US organisation, late 2025
Per Google, the actor signed in with stolen Super Admin credentials through a US residential proxy close to the victim's location, and ran an MFA push-fatigue attack until the user accepted a prompt. It then registered a device model never before seen in the tenant, and added passkeys, MFA backup codes and rogue MFA devices for persistence.
With Super Admin rights it mass-shared files from the organisation's cloud storage to an actor-controlled email account, downloaded files directly, exported full employee lists and started a domain-wide Takeout export (Google's admin feature for exporting an organisation's data). It also covered its tracks: it queried the security investigation and email audit logs to see what defenders could see, and viewed, modified and deleted Alert Center notifications.
Google says the actor stayed for several months, syncing mailboxes to external email clients through OAuth and signing back in with passkeys that survived password and session resets. In at least one case, Google says, the victim reset passwords but did not revoke existing sessions, so the actor kept access. After it was removed, the actor used email accounts impersonating Victim 1's employees to target their colleagues and professional contacts.
Victim 2: a second US organisation, 2026
Victim 2 used Microsoft Entra ID as the single sign-on for its Google Workspace. Per Google, the actor impersonated an employee of Victim 1 on an encrypted messaging app to build trust with a target, then shared phishing links, framed as a document about a technical issue, that harvested Entra ID credentials and authentication tokens. After signing in it registered a new MFA device in the SSO system and bulk-downloaded files from the user's cloud drive. The messaging profile was later changed to a completely different persona, which Google believes was used to go after other organisations in the same professional circles.
That is the pattern to take away: every compromised organisation becomes a source of believable pretexts for the next one. Rosters, contact lists, CVs, event schedules and real email threads make the next approach look like a known colleague. Google says the group also puts a reference to a previous victim or the lure topic in its phishing URLs.
Google notes that one figure in its post, showing the social-engineering conversation, is an "LLM-generated depiction" based on the observed activity, not a capture of the real messages.
Useful next, if this was relevant to you
Timeline
| Date | Event | Source |
|---|---|---|
| Since 2015 | GTIG tracks and disrupts the group's phishing campaigns | |
| Late 2025 | Victim 1 (US) Google Workspace compromised; access kept for several months | |
| June 2026 | A COLD RELIC redirector URL referencing Victim 1 is submitted to VirusTotal | |
| 2026 | Victim 2 (US) compromised through an employee impersonating Victim 1 staff; Entra ID credentials and tokens phished | |
| Nov 2026 | Google observes new COLD RELIC phishing operations | |
| 17 Nov 2026 | GTIG and Mandiant publish "COLD RELIC: A Case Study in Impersonation and Compromise" | |
| 18 Nov 2026 | The Hacker News reports the case study | The Hacker News |
What security leaders should do this week
- Retire push, SMS and voice MFA for admins first. Google's first recommendation is phishing-resistant MFA (FIDO2/WebAuthn security keys). Super Admin and Global Admin accounts go first, then executives, and staff who deal with government, defence or policy contacts. Google also suggests its Advanced Protection Program for high-risk users.
- Control who can add an authenticator. Require an existing admin or an out-of-band check to approve a new MFA device, passkey or backup codes, and alert on every passkey or device enrolment on an admin account, especially just after a burst of MFA prompts.
- Fix the reset playbook. A password reset alone did not remove this actor. Your account-compromise and business email compromise playbooks should revoke browser sessions, refresh tokens and third-party OAuth grants in both Workspace and Entra ID, and remove unrecognised passkeys, backup codes and MFA devices, every time.
- Audit OAuth mail access. Look for OAuth grants of full mail scope to clients you have not approved, such as desktop mail clients and external mail services, which Google says the actor used to sync inboxes.
- Alert on the admin actions an attacker needs. In Google Workspace these include a Customer Takeout export being created, a user list download, investigation-tool queries, batch deletion of Alert Center alerts (especially government-backed attack alerts), bulk Drive downloads and spikes in external sharing. Google's post maps each to an event name and ATT&CK technique.
- Treat residential-proxy sign-ins to admin accounts as suspicious. Google advises restricting admin access from consumer ISPs, VPNs and residential proxy networks and enriching sign-in logs with proxy and VPN reputation. Decide whether you can enforce that with conditional access.
- Limit Takeout and mass sharing. Decide who may use Google Takeout, and use data protection rules to warn on or block mass external sharing of files that look like rosters, credentials or mail archives.
- Brief the people who will be impersonated. Staff whose names, roles and contacts could have been stolen from a partner should know the pattern: a known colleague moves the conversation to a messaging app and sends a "document" link. Verify by a second channel before signing in anywhere.
Put it to work
- Phishing-resistant MFA planner: sequence the move to security keys and passkeys.
- MFA bypass lab: see push fatigue and token theft from the attacker's side.
- Identity compromise response checklist: the session, token and authenticator clean-up Google says was missed.
- OAuth consent lab: practise spotting OAuth grants that give a third party your mail.
- BEC response checklist: when the impersonated mailbox or the impersonated partner is yours.
FAQ
Who is COLD RELIC?
COLD RELIC is Google's current name for the group it previously tracked as COLDRIVER and UNC4057, also known as Star Blizzard, Callisto and Gossamer Bear. Google calls it a Russia-nexus, state-sponsored cyber espionage group and says it has tracked and disrupted its phishing campaigns since 2015. Google cites the US Department of Justice's assessment that the group operates within Russia's Federal Security Service (FSB).
What did COLD RELIC do with stolen cloud data?
Per Google, after compromising a US organisation's Google Workspace in late 2025, the group used what it took (employee lists, emails, contacts) to impersonate that organisation's employees. In 2026 it posed as one of them on an encrypted messaging app to phish a person at a second US organisation, stole their Microsoft Entra ID credentials and tokens, and downloaded files from their cloud drive.
Which organisations were breached by COLD RELIC?
Google does not name either victim. It describes both as US-based organisations and does not give their sectors. The group's usual targets, per Google, are government officials, defence contractors and NGOs in Europe and North America. Any name attached to this report elsewhere is not from Google's post.
How did COLD RELIC get past MFA?
In the first case Google says the actor signed in with stolen Super Admin credentials through a US residential proxy and ran an MFA push-fatigue attack until the user accepted a prompt. It then registered its own device, passkeys and backup codes. In the second case it phished Entra ID credentials and tokens and registered a new MFA device. Google recommends FIDO2 security keys in place of push, SMS and voice MFA.
Why did resetting passwords not remove COLD RELIC?
Google says that in at least one case the victim reset passwords but did not revoke existing sessions, so the actor kept access, and that passkeys the actor had enrolled survived password and session resets. Google's advice is to revoke sessions, refresh tokens and third-party OAuth grants, and remove unauthorised passkeys, backup codes and MFA devices, every time you reset a compromised account.
Is COLD RELIC still active?
Yes, according to Google: "GTIG has observed evidence of new phishing operations as recently as November 2026."
Was Google Workspace hacked?
No. Google describes the abuse of legitimate accounts and admin features (stolen credentials, MFA fatigue, Takeout exports, OAuth mail sync) in its customers' tenants, not a flaw in or breach of Google Workspace or Microsoft Entra ID.
Sources
- Google Threat Intelligence Group / Mandiant: COLD RELIC: A Case Study in Impersonation and Compromise (17 Nov 2026; Josh Murchie, John Wolfram, Joseph Spero, Peter Ukhanov, Fernando Tomlinson Jr., Wesley Shields). Primary source, including the detection table and recommendations.
- Google Threat Intelligence Group: COLDRIVER using new malware to steal documents from Western targets and NGOs (earlier reporting on the same group, under its old name)
- The Hacker News: Russian COLD RELIC hackers use stolen Google Workspace data to breach second U.S. organization (18 Nov 2026)
- Google Workspace Admin Help: allow or block Google Takeout
This article is educational guidance for security leaders, not a threat-intelligence report. Indicators and detection rules are in Google's post; check it for updates before acting.