COLD RELIC turns one cloud breach into wider espionage, Mandiant finds
Mandiant has linked two intrusions at US organisations, showing how the Russian group COLD RELIC reuses stolen identities and data to reach new targets.

Key takeaways
- Mandiant linked a late-2025 breach at one US organisation to a 2026 intrusion at another, saying stolen identities and data helped the second attack.
- The group used a US residential proxy, a stolen Super Admin account and repeated MFA prompts to enter a cloud productivity tenant.
- Attackers kept access with passkeys and rogue MFA devices, and tried to erase evidence by deleting Alert Center notifications.
- Administrators can disable Google Takeout organisation-wide and should watch for Customer Takeout events in audit logs.
In this article
What Mandiant found
Mandiant, Google's incident response unit, has published a case study of linked intrusions by the Russian espionage group COLD RELIC. According to the Google Cloud Blog, the group broke into one US organisation in late 2025 and then used material taken from that breach to attack a second US organisation in 2026.
The group is also tracked as COLDRIVER, UNC4057, Star Blizzard, Callisto and Gossamer Bear. The US Department of Justice has assessed that it operates within Russia's Federal Security Service, the FSB. Mandiant says the group targets government officials, defence contractors and non-governmental organisations in Europe and North America.
How the first intrusion unfolded
In the late-2025 case, the attackers reached the victim's cloud productivity environment through a residential proxy network in the US, placed close to the victim's location. They signed in with stolen credentials for a Super Admin account, a role with wide control over the tenant.
Mandiant reports that the attackers combined those credentials with a push notification fatigue attack. This means sending repeated multi-factor authentication prompts until a user approves one. After a request was accepted, a device model that had not appeared before in the organisation's tenant was registered.
The group then used passkeys, backup codes for multi-factor authentication and its own registered devices to keep access. With Super Admin rights, it copied the organisation's cloud storage by sharing files to an email account it controlled and by downloading files directly. It also exported employee lists and started a tenant-wide export through Google Workspace Customer Takeout, an administrative tool for exporting all data in a domain.
Covering tracks and staying in place
While inside the tenant, the intruders opened the Admin Console and checked security investigation and email audit logs to see what defenders could observe. They also viewed, changed and deleted Alert Center notifications, which Mandiant describes as an attempt to remove evidence.
Access lasted several months. Mandiant lists further activity: syncing email inboxes to outside email clients through OAuth, signing in again with attacker-enrolled passkeys that survived password and session resets, and opening specific documents about credentials, email archives and personnel.
When the group was removed, it set up email accounts that posed as employees of the victim and used them to approach other people. Mandiant says the 2026 intrusion at the second organisation showed signs that relationship details from the first breach helped the attackers choose their targets and pretexts.
The phishing method behind the links

COLD RELIC commonly sends a link to a compromised website, which then redirects the visitor to the real phishing domain. This hides the phishing address and lets the attackers fingerprint visitors to decide whether to redirect them.
The links usually carry a parameter whose value is set by the attackers. Mandiant says this value often points to an earlier victim or to the theme of the lure. In the 2026 investigation, responders found that the group first made contact through an encrypted messaging service, posing as a staff member of the first victim. Separately, Google Threat Intelligence Group matched a redirector URL submitted to VirusTotal in June 2026 whose query parameter named the first victim organisation.
What organisations should watch
Mandiant says COLD RELIC remains active and that Google's threat intelligence team saw signs of new phishing operations as recently as November 2026. Google issues government-backed attack alerts when it detects the group targeting Google accounts.
The case study points to controls that matter in cloud environments: limiting Super Admin accounts, resisting repeated authentication prompts, reviewing enrolled devices and passkeys, and monitoring administrative exports. Mandiant notes that Customer Takeout requests appear in Google Workspace Admin audit logs under CUSTOMER_TAKEOUT_CREATED, and that administrators can turn off Google Takeout for users across the organisation. Organisations should follow the detection and hardening guidance in the full Mandiant report and their cloud provider's official security documentation.
Sources
- COLD RELIC: A Case Study in Impersonation and CompromiseGoogle Cloud Blog
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



