What happened
During a late-2025 intrusion, COLD RELIC compromised the cloud productivity environment of a US-based organisation and retained access for several months. In the first intrusion, COLD RELIC used stolen Super Admin credentials with MFA push fatigue, then enrolled passkeys, backup codes and a previously unseen MFA device. The actor downloaded the organisation’s cloud drive, exported employee lists, initiated Google Workspace Customer Takeout and modified or deleted Alert Center notifications.
In 2026, Mandiant investigated a second US-based organisation where stolen context from the first victim supported impersonation and credential phishing. In the second intrusion, COLD RELIC impersonated an employee over an encrypted messaging application, stole Microsoft Entra ID credentials and tokens, registered a new MFA device and bulk-downloaded accessible cloud files. The second organisation used Entra ID as its federated identity provider for Google Workspace.
Google published the linked case study on November 17, 2026 and said it had observed new COLD RELIC phishing operations as recently as November 2026. Google’s published detection opportunities include CUSTOMER_TAKEOUT_CREATED, Batch Delete Operation Performed, User List Download, Passkey enrolled, Device Registered and broad mail OAuth Authorize events.
The cited report did not publish the victim names or current campaign IP addresses and domains. Attribution posture: Google identifies COLD RELIC as Russian state-sponsored and cites a US Department of Justice assessment that the group operates within Russia’s FSB.
Why this matters now
The campaign shows that cloud compromise creates two victim populations: the breached tenant and every trusted person or organisation represented in its data. Employee directories, correspondence, event schedules and documents become high-quality social-engineering material that can defeat familiarity-based verification at partners and suppliers.
The persistence mechanisms defeat partial containment. Password changes and session resets did not remove attacker-enrolled passkeys, backup codes, rogue MFA devices or broad OAuth access in the reported case. Organisations need a single cloud-identity containment procedure spanning authenticators, sessions, tokens, federated identity and third-party applications.
COLD RELIC also operated across Google Workspace and Microsoft Entra ID, using the federation boundary rather than respecting platform ownership. Separate SaaS administration teams can miss the full incident if one team closes its account while another leaves the identity provider, OAuth grant or enrolled device intact.
The decision for security leaders
Direct the identity team to test containment as a complete sequence rather than a collection of optional actions. Password reset, session revocation, authenticator removal, backup-code invalidation, OAuth revocation and federated-identity review must be executed and evidenced together for compromised cloud accounts.
Make cloud audit events part of executive incident thresholds. Tenant-wide export creation, privileged directory download, security-alert deletion or a new authenticator after MFA fatigue should trigger immediate investigation even when source geography appears normal, because COLD RELIC used residential proxies near the victim.
Extend incident communications to trusted external relationships when stolen correspondence or directories can support impersonation. The affected organisation may need to warn selected partners through established channels without overstating compromise at those partners or disclosing unnecessary victim information.
Evidence of closure
- Privileged-account tests generate alerts for MFA fatigue followed by successful authentication.
- Cloud audit rules detect Customer Takeout, directory exports, alert deletion and new authenticator enrolment.
- The containment runbook requires session, token, OAuth and authenticator revocation in one approved workflow.
- Cloud-storage scans show no unresolved credentials or email archives in user-accessible locations.
The Security.io assessment
The case study is significant because it is based on linked Mandiant incident-response investigations rather than generic phishing telemetry. It demonstrates observed cross-cloud persistence, defensive tampering, mass collection and downstream impersonation. The November disclosure provides concrete audit events that enterprises can operationalise immediately.
The cited report did not publish the victim names or current campaign IP addresses and domains. That prevents a conventional IOC-led hunt and places greater weight on identity behaviour, administrative audit events, OAuth activity and unusual use of trusted corporate context. Organisations should not interpret the absence of published infrastructure as reduced campaign activity.
The strategic lesson is that cloud data has offensive value beyond confidentiality loss. Directories, archives and relationship history can become access material against other organisations. This expands incident scope from the compromised tenant to the identities and relationships represented inside it, while still requiring careful separation between confirmed compromise and downstream targeting.
Questions for the morning meeting
- Do cloud-account containment procedures revoke sessions, OAuth grants, passkeys and backup codes together?
- Can the SOC detect tenant-wide exports, directory downloads and alert deletion by privileged users?
- Which credentials, archives and personnel records remain stored in user-accessible cloud drives?
- How are partners warned when stolen corporate context may support impersonation?