The COLD RELIC cloud compromise described by Google Threat Intelligence Group links two US organizations across separate incidents. In late 2025, the actor stole Workspace data from the first organization. Google assesses that stolen relationship context helped the actor approach a second organization in 2026. A third-party risk management platform can help assess supplier exposure when that relationship matters.
The second organization used Microsoft Entra ID for federated sign-on to Google Workspace. Google confirmed an Entra ID account compromise, a new MFA device and bulk downloads of accessible cloud files. Responders should follow identity access, authenticator changes and file activity across both systems.
Why Does the COLD RELIC Cloud Compromise Require a Cross-Cloud Playbook?
Google attributes COLD RELIC to the actor also tracked as Star Blizzard, COLDRIVER and UNC4057. Its November 17, 2026 report separates a late-2025 Workspace intrusion from a 2026 approach against another organization. The reported connection is the actor’s use of an impersonated employee from the first organization, not a shared tenant or simultaneous login.
At the first organization, Google observed stolen Super Admin credentials, repeated MFA prompts and an approved push. The actor enrolled a device, used passkeys and backup codes, and collected tenant data. In 2026, a phishing conversation aimed at Entra ID credentials and tokens preceded the second account compromise.
Step 1: Join Entra ID and Workspace Records
Build a timeline for the affected user before changing the account. Join Entra ID sign-ins, MFA registration and session records with Workspace login, OAuth, Admin and Drive audit events. Record event time, account, device and network context so that a federated sign-in can be compared with later Workspace access.
Check whether an unfamiliar sign-in preceded a new authenticator and a burst of file downloads. Google’s case includes a newly registered MFA device in the SSO solution and bulk downloads from files accessible to that user. Look for repeated MFA challenges followed by success; Google maps this pattern to ATT&CK T1621.
Do not clear a sign-in solely because its location looks familiar. Google observed residential proxies near users’ usual locations, which can defeat impossible-travel alerts. Compare device history and network reputation with the user’s normal access before deciding that an event is routine.
Step 2: Revoke Sessions and Remove Rogue Authenticators
Contain the affected identity in both Entra ID and Workspace. Reset credentials, revoke active browser sessions and refresh tokens, and withdraw unauthorized third-party OAuth permissions. Then remove unrecognized passkeys, backup codes and MFA devices before verified re-enrollment.
Check the revocation and registration audit records for each system, then test whether old sessions still work. Google reports one investigated intrusion where password resets failed because existing sessions were not revoked. Its case also describes attacker-enrolled passkeys persisting across resets, making authenticator review essential.
Preserve the event trail while containing access: note who approved each new device and when. Google maps device registration to T1098.005 and backup-code authentication to T1078.004. Require identity verification outside the compromised channel before approving replacement authenticators.
Step 3: Trace What the Account Collected and Exported
Review Workspace Drive Download events and sharing-permission changes for the affected account. Identify file names, owners, download volume and external sharing changes in the response record. Compare the time window with Entra ID access and new MFA registration to determine what the compromised user could reach.
For privileged accounts, also inspect Workspace Admin events for Customer Takeout creation and user-list downloads. Google observed both in the first organization, along with mass sharing and direct downloads. Its detection mapping associates automated collection with T1119, cloud file access with T1530 and external sharing with T1567.002.
Review OAuth Authorize events for broad mail access by unexpected clients. Google observed external email-client syncing in the first intrusion and maps this collection and token abuse to T1114.002 and T1528. The second case confirms bulk file downloads; do not assume its account performed every action seen in the first case.
Step 4: Review Admin Changes and Credential Exposure
Search Workspace Admin logs for Alert Center changes, alert deletion, user-list exports and investigation queries. Google’s first case included deleted security alerts and queries of defensive logs. Its detection guidance points to Batch Delete Operation Performed and Investigation Query events; these can show where visibility was impaired.
Check Drive search and access events for documents containing credentials or personnel information. Google says the actor sought such material and maps directory discovery to T1087.004, audit querying to T1082, and sensitive-file search to T1083 and T1552.001. Treat each mapping as a lead for review, not proof that every technique occurred in the second organization.
Brandefense Blackhole recorded 2,659 dark-chat posts matching the broad term “COLD” from August 20 through November 18, 2026. The unit is matching posts, not distinct actors, victims or mentions of this case. This count cannot establish COLD RELIC activity or confirm either intrusion.
Step 5: Protect Contacts and Prevent Repeat Access
Review which contact lists, rosters and relationship documents were accessed or exported. Google assesses that stolen context aided the later impersonation, and it observed the actor pose as an employee of the first organization. Tell relevant contact owners how to verify unexpected requests through a known channel.
For privileged accounts, require approval for new authenticators and prefer phishing-resistant MFA. Review MFA fatigue alerts and unexpected OAuth grants as continuing detection priorities.
- Join the timelineCompare Entra ID sign-ins and MFA changes with Workspace login and Drive events.
- Contain identityRevoke sessions, refresh tokens and unauthorized OAuth grants in both systems.
- Remove persistenceAudit passkeys, backup codes and device registrations before re-enrollment.
- Scope collectionReview Drive downloads, external sharing, Customer Takeout and directory exports.
- Protect contactsVerify suspicious relationship-based requests through known channels.
This order links the initial identity event to the data it enabled, then closes persistence and limits further exposure. Google’s November 2026 observation of new phishing activity supports continued monitoring. It does not mean every similar message is attributable to this actor.
Frequently Asked Questions
Did Google describe one intrusion or two?
Two linked intrusions at separate US organizations. The first occurred in late 2025; Google investigated the second in 2026.
Why is a password reset insufficient here?
Google observed an intrusion where active sessions survived a password reset. Responders should also revoke sessions, refresh tokens and OAuth grants, then remove unauthorized authenticators.
What evidence connects Entra ID access with Workspace downloads?
Google reports that the second organization federated Google Workspace sign-on through Entra ID. Review Entra ID sign-ins and MFA registration alongside Workspace file access and download logs.
