Cold Relic Used Stolen Workspace Data to Target a Second U.S. Organization
Google says the Russia-linked espionage group Cold Relic used information stolen from one U.S. organization to impersonate its employees and compromise an account at a second U.S. organization. The activity shows how a cloud account takeover can extend beyond the original victim when attackers exploit trusted business relationships and federated identity systems.
Google’s Mandiant and Threat Intelligence Group (GTIG) investigated the incidents. The first organization’s Google Workspace environment was compromised in late 2025; a separate intrusion at another U.S.-based organization followed in 2026. Google tracks the group as Cold Relic, also known as COLDRIVER, UNC4057, Callisto, Gossamer Bear and Star Blizzard.
Workspace administrator account enabled broad access
In the late-2025 incident, the attackers accessed a Super Admin account using stolen credentials and an MFA push notification fatigue attack. Google said the initial access came through a residential proxy network located close to the victim. The account holder accepted an authentication request, after which the attackers enrolled their own device in the organization’s tenant. Google also observed the use of passkeys and MFA backup codes as alternate access methods.
With administrator privileges, Cold Relic collected data from the organization’s cloud environment. Google reported that the group mass-shared files to an attacker-operated email account and downloaded files directly. The attackers also used Google Takeout to export employee lists and organization-wide data.
The intruders accessed the Admin Console and queried security investigation and email audit logs, apparently checking what activity defenders could see. They also viewed, changed and deleted Alert Center notifications, actions Google said removed forensic evidence related to the initial intrusion.
Access continued for several months. During that time, the group synchronized email inboxes to external email clients through OAuth and reused attacker-enrolled passkeys that had not been revoked. Google said those passkeys remained usable across password and session resets and bypassed device registration controls. The attackers also sought documents concerning credentials, email archives and personnel information.
Stolen identities became a route to another victim
After the first organization removed the attackers, Google observed them using email accounts that impersonated its employees to approach internal staff and professional associates. The stolen employee information and messages helped the group create plausible pretexts and present itself as a known contact.
The campaign then reached a second U.S. organization, which used Microsoft Entra ID as its federated single sign-on provider for Google Workspace. Cold Relic impersonated an employee of the first victim and contacted a target through a secure messaging application. The attacker sent a link presented as a document about a technical issue; Google said the link was designed to steal Entra ID credentials and authentication tokens.
After the target authenticated with stolen credentials, Cold Relic enrolled and used a new MFA device in the SSO solution to maintain access to the account. The group then bulk-downloaded files available in the compromised user’s cloud drive. Google reported that the messaging account used to begin the conversation later changed its profile name to another persona, which may have been intended to approach additional organizations in the same professional network.
Identity and monitoring controls are central to response
The incidents highlight risks that span cloud storage, identity providers, authentication methods and human relationships. In particular, an MFA prompt accepted by a user, combined with attacker enrollment of an authentication device, can give an intruder a foothold that outlasts routine password or session changes. The use of a federated identity provider also means that activity in one identity environment can have consequences for services connected to it.
Organizations should review cloud identity and administrator activity for unexpected device, passkey or MFA-method enrollment, suspicious OAuth access, unusual bulk downloads and changes to alerting or audit visibility. Security teams should verify that revoked credentials and authentication factors cannot be reused, and ensure incident response procedures cover both the identity provider and connected cloud applications.
Google’s account also underscores the value of treating data theft as a possible precursor to further social engineering. Employee directories, email archives and operational documents can help an attacker convincingly impersonate a trusted colleague. Organizations should warn employees and external partners when an account or correspondence has been compromised, and assess whether contacts in shared professional networks may also have been targeted.