Back to Intelligence

COLD RELIC (COLDRIVER/UNC4057) Weaponizes Stolen Google Workspace Data for Third-Party Compromise — Detection and Response Guide

SA
Security Arsenal Team
November 18, 2026
11 min read

Google's Mandiant and Threat Intelligence Group (GTIG) have published findings from a 2026 intrusion investigation that should concern every organization running Google Workspace — and every organization that does business with one that does. The Russia-linked espionage group tracked as COLD RELIC (formerly COLDRIVER and UNC4057) was observed weaponizing a single cloud compromise into a launchpad for a series of linked intrusions, culminating in the breach of a second, unrelated U.S.-based organization via stolen Google Workspace data.

This is the supply-chain-of-trust problem in its purest form. The initial victim was not the endgame. COLD RELIC harvested data from the first organization's Google Workspace tenant — email, files, contacts, and collaboration artifacts — and then leveraged that material (stolen credentials, trusted relationship context, and sensitive correspondence) to stage a convincing follow-on intrusion against a downstream organization. One tenant's breach became the reconnaissance and access vector for the next.

No CVE is at the center of this campaign. There is no patch that fixes this. The defensive lesson is architectural: identity is the perimeter, cloud tenant telemetry is your only ground truth, and third-party trust relationships are now an exploitation surface. If your SOC treats a Google Workspace compromise as a contained incident limited to one tenant, this campaign demonstrates exactly why that assumption will cost you — or your partners — dearly.

Technical Analysis

Threat Actor Profile

COLD RELIC is a Russia-nexus cyber espionage group with a long operational history under the names COLDRIVER and UNC4057. The group has historically targeted NGOs, think tanks, journalists, government entities, and defense-adjacent organizations, primarily through credential phishing, adversary-in-the-middle (AiTM) phishing frameworks, and abuse of OAuth consent flows to gain durable access to cloud email. The 2026 campaign documented by Mandiant/GTIG represents a maturation of their playbook: pivoting from single-target espionage to multi-hop compromise chains built on harvested cloud data.

Attack Chain (Defender's Perspective)

Based on the reported investigation, the intrusion sequence follows this pattern:

  1. Initial access to Organization A's Google Workspace tenant — consistent with COLD RELIC's established tradecraft: targeted credential phishing or OAuth consent abuse, often against high-value individuals (executives, IT admins, or staff with access to sensitive partner correspondence).
  2. Data harvesting — once inside, the actor mines Gmail and Google Drive for high-value material: credentials, VPN details, partner contact lists, contract documents, technical architecture diagrams, and prior correspondence that reveals trusted communication channels with Organization B.
  3. Weaponization of stolen context — the actor uses the harvested data to craft highly credible pretexts against Organization B. Because the phishing or social engineering originates with legitimate context (real names, real projects, real document formats, sometimes real reply-thread hijacking), conventional phishing defenses and user heuristics fail.
  4. Breach of Organization B — a technically distinct victim, compromised not because its own perimeter was weak, but because its trust in Organization A was exploitable.

Why This Is Dangerous

  • No malware signature catches this. The techniques — OAuth token use, legitimate API access via the Google Workspace Admin SDK and Gmail API, and social engineering built on authentic context — all generate traffic indistinguishable from normal user behavior unless you are specifically hunting for it.
  • The blast radius is invisible. When a tenant is breached, the exposed data defines a second, hidden victim set: every partner, customer, and correspondent represented in that data. Incident response scoping that stops at "which accounts were accessed" misses "whose data about whom was exfiltrated."
  • Third-party risk becomes real-time risk. Organization B in this campaign was breached through no failure of its own patch management. Traditional vendor risk questionnaires would not have caught this.

Exploitation Status

  • Active exploitation confirmed: This is a documented, investigated intrusion campaign from 2026, not a theoretical threat.
  • No CVE / CISA KEV entry: This is a technique-driven campaign (credential theft, OAuth abuse, social engineering, and data weaponization), not a software vulnerability.
  • Attribution: Russia-nexus espionage; tracked by Google as COLD RELIC, formerly COLDRIVER / UNC4057.

Detection & Response

The detections below target the observable behaviors this campaign depends on: anomalous OAuth grants and token use in Google Workspace, suspicious session and access patterns against Gmail/Drive, and endpoint artifacts associated with browser credential theft that often precede cloud account takeover. Tune thresholds to your environment's baseline — an OAuth grant by a user in a multinational org looks different from one in a 50-person firm.

Sigma Rules

YAML
---
title: Suspicious OAuth Application Consent Grant in Google Workspace
id: 3f9a2c71-8b4d-4e5f-9a6c-2d1e0b8f7a53
status: experimental
description: Detects OAuth consent grants to third-party applications with high-risk scopes (Gmail read, Drive full access) consistent with COLD RELIC / COLDRIVER tradecraft for persistent mailbox and file access.
references:
  - https://thehackernews.com/2026/11/russian-cold-relic-hackers-use-stolen.html
  - https://attack.mitre.org/techniques/T1550/001/
  - https://attack.mitre.org/techniques/T1528/
author: Security Arsenal
date: 2026/11/15
tags:
  - attack.persistence
  - attack.credential_access
  - attack.t1550.001
logsource:
  product: google_workspace
  service: token
detection:
  selection_scopes:
    scope|contains:
      - 'mail.google.com'
      - 'gmail.readonly'
      - 'gmail.modify'
      - 'drive.readonly'
      - 'drive'
  selection_event:
    event_name: 'authorize'
  filter_known_apps:
    app_name|contains:
      - 'Google'
      - 'Zoom'
      - 'Slack'
      - 'Microsoft'
  condition: selection_scopes and selection_event and not filter_known_apps
falsepositives:
  - Legitimate new SaaS integrations authorized by users
  - IT-approved third-party backup or migration tools
level: high
---
title: Google Workspace Login from High-Risk or Anonymous Infrastructure
id: 8c1d4e62-7a3b-4f09-b2e5-9d6c1a4f8e27
status: experimental
description: Detects successful Google Workspace logins originating from VPN, proxy, Tor, or hosting-provider IP ranges, a common indicator of AiTM phishing session replay or remote access by COLD RELIC operators.
references:
  - https://thehackernews.com/2026/11/russian-cold-relic-hackers-use-stolen.html
  - https://attack.mitre.org/techniques/T1078/004/
author: Security Arsenal
date: 2026/11/15
tags:
  - attack.initial_access
  - attack.t1078.004
logsource:
  product: google_workspace
  service: login
detection:
  selection:
    event_name: 'login_success'
  condition_risky:
    is_suspicious: 'true'
  filter:
    ip_address|cidr:
      - '10.0.0.0/8'
      - '172.16.0.0/12'
      - '192.168.0.0/16'
  condition: selection and (condition_risky or not filter)
falsepositives:
  - Corporate VPN egress IPs (tune to known egress ranges)
  - Traveling executives on hotel or mobile networks
level: medium
---
title: Bulk Gmail Export or Drive Mass Download Activity
id: 5e7b9f30-2c8d-4a16-9f43-7e2b5d0c6a91
status: experimental
description: Detects Google Takeout exports, mass Drive downloads, or Gmail data export events consistent with the data harvesting phase observed in COLD RELIC intrusions prior to downstream targeting.
references:
  - https://thehackernews.com/2026/11/russian-cold-relic-hackers-use-stolen.html
  - https://attack.mitre.org/techniques/T1530/
  - https://attack.mitre.org/techniques/T1567/002/
author: Security Arsenal
date: 2026/11/15
tags:
  - attack.collection
  - attack.exfiltration
  - attack.t1530
logsource:
  product: google_workspace
  service: drive
detection:
  selection:
    event_name:
      - 'download'
      - 'export'
  condition: selection
falsepositives:
  - Users performing legitimate large migrations or backups
  - Automated archival jobs (exclude service accounts)
level: medium

KQL — Microsoft Sentinel

The query below hunts Google Workspace audit logs ingested into Sentinel (via the Google Workspace connector or CEF/Syslog) for the convergence of behaviors this campaign exhibits: new OAuth grants with mail/Drive scopes, logins flagged as suspicious, and subsequent bulk access events against the same accounts.

KQL — Microsoft Sentinel / Defender
// Hunt: COLD RELIC-style Google Workspace account takeover and data harvesting
// Requires Google Workspace audit logs ingested into Sentinel (table name may vary by connector)
let lookback = 14d;
let SuspiciousLogins = GWorkspaceActivity
| where TimeGenerated > ago(lookback)
| where EventName =~ "login_success"
| where IsSuspicious == true or IpAddress !startswith "10."
| project LoginTime=TimeGenerated, ActorEmail, LoginIp=IpAddress, LoginGeo=Region;
let RiskyOAuth = GWorkspaceActivity
| where TimeGenerated > ago(lookback)
| where EventName =~ "authorize"
| where ScopeData has_any ("mail.google.com", "gmail.readonly", "gmail.modify", "drive")
| where AppName !has_any ("Google", "Slack", "Zoom", "Microsoft")
| project GrantTime=TimeGenerated, ActorEmail, AppName, ScopeData;
let BulkAccess = GWorkspaceActivity
| where TimeGenerated > ago(lookback)
| where EventName in ("download", "export", "view")
| summarize AccessCount=count(), FirstAccess=min(TimeGenerated), LastAccess=max(TimeGenerated) by ActorEmail, bin(TimeGenerated, 1h)
| where AccessCount > 200;
SuspiciousLogins
| join kind=inner (RiskyOAuth) on ActorEmail
| join kind=inner (BulkAccess) on ActorEmail
| project ActorEmail, LoginTime, LoginIp, LoginGeo, GrantTime, AppName, ScopeData, AccessCount, FirstAccess, LastAccess
| sort by ActorEmail asc;

For organizations without native Google Workspace ingestion, hunt the endpoint side instead — credential theft against browser stores frequently precedes the cloud phase:

KQL — Microsoft Sentinel / Defender
// Endpoint hunt: processes accessing browser credential stores (precursor to session theft)
DeviceProcessEvents
| where TimeGenerated > ago(7d)
| where ProcessCommandLine has_any ("Login Data", "Cookies", "Web Data", "Local State")
   or FileName in~ ("sqlite3.exe", "esentutl.exe", "rundll32.exe")
      and ProcessCommandLine has_any ("\\AppData\\Local\\Google\\Chrome", "\\AppData\\Local\\Microsoft\\Edge")
| where InitiatingProcessFileName !in~ ("chrome.exe", "msedge.exe", "msmpeng.exe")
| project TimeGenerated, DeviceName, AccountName, FileName, ProcessCommandLine, InitiatingProcessFileName
| sort by TimeGenerated desc;

Velociraptor VQL

This artifact hunts endpoints for processes with handles or access patterns against browser credential stores — the on-host artifact of the session-token theft that commonly seeds these cloud intrusions.

VQL — Velociraptor
-- Hunt for non-browser processes reading Chrome/Edge credential and cookie stores
SELECT Pid,
       Name,
       Exe,
       CommandLine,
       Username,
       CreateTime
FROM pslist()
WHERE CommandLine =~ '(?i)(Login Data|Cookies|Local State|Web Data)'
  AND Name !~ '(?i)^(chrome|msedge|msmpeng|svchost)\\.exe$'
  AND Exe !~ '(?i)(Google\\\\Chrome|Microsoft\\\\Edge)'

For network-level confirmation, pair with a netstat artifact correlating unusual outbound sessions from browsers or scripting hosts to non-standard destinations:

VQL — Velociraptor
-- Correlate browser processes with unexpected remote endpoints
SELECT Pid,
       Name,
       Exe,
       'Laddr.IP' AS LocalIP,
       'Laddr.Port' AS LocalPort,
       'Raddr.IP' AS RemoteIP,
       'Raddr.Port' AS RemotePort,
       Status
FROM netstat()
WHERE Name =~ '(?i)(chrome|msedge|powershell|wscript|mshta|rundll32)\\.exe$'
  AND 'Raddr.IP' !~ '^(10\\.|172\\.(1[6-9]|2[0-9]|3[01])\\.|192\\.168\\.|127\\.)'
  AND RemotePort in (80, 443, 8080, 8443)

Remediation & Hardening Script

The PowerShell below audits a Google Workspace tenant (via GAM, the standard admin tooling) for the exact artifacts this campaign abuses: risky third-party OAuth grants, suspicious forwarding rules, and accounts without 2SV. Run it from an admin workstation with GAM configured against your tenant.

PowerShell
# COLD RELIC response: Google Workspace tenant audit and hardening
# Requires GAM (https://github.com/GAM-team/GAM) authenticated as super admin

$gam = "C:\GAM\gam.exe"
$reportDir = "C:\IR\GWS_Audit_$(Get-Date -Format 'yyyyMMdd')"
New-Item -ItemType Directory -Path $reportDir -Force | Out-Null

Write-Host "[1/5] Enumerating all third-party OAuth token grants..." -ForegroundColor Cyan
& $gam all users print tokens > "$reportDir\oauth_tokens.csv"
# Flag grants with mail or drive scopes
Import-Csv "$reportDir\oauth_tokens.csv" |
  Where-Object { $_.scopes -match 'mail.google.com|gmail|drive' -and
                 $_.clientId -notmatch 'apps.googleusercontent.com' } |
  Export-Csv "$reportDir\oauth_high_risk.csv" -NoTypeInformation

Write-Host "[2/5] Checking for suspicious Gmail forwarding rules (exfil channel)..." -ForegroundColor Cyan
& $gam all users print forward > "$reportDir\forwarding.csv"
Import-Csv "$reportDir\forwarding.csv" |
  Where-Object { $_.forwardTo -notmatch '@yourdomain\.com$' } |
  Export-Csv "$reportDir\forwarding_external.csv" -NoTypeInformation

Write-Host "[3/5] Identifying accounts without 2-Step Verification..." -ForegroundColor Cyan
& $gam report users parameters accounts:is_2sv_enrolled > "$reportDir\2sv_status.csv"

Write-Host "[4/5] Checking for app passwords in use (bypasses 2SV)..." -ForegroundColor Cyan
& $gam all users show asps > "$reportDir\app_passwords.csv"

Write-Host "[5/5] Pulling recent admin audit events for review..." -ForegroundColor Cyan
& $gam report admin start "-7d" > "$reportDir\admin_audit.csv"

Write-Host "`nAudit complete. Review flagged files in $reportDir" -ForegroundColor Green
Write-Host "Priority actions:"
Write-Host " - Revoke tokens in oauth_high_risk.csv:  gam user <email> deprovision"
Write-Host " - Remove external forwarding rules:      gam user <email> delete forward <address>"
Write-Host " - Enforce 2SV org-wide via Admin Console > Security > 2-Step Verification"
Write-Host " - Reset credentials for any account with anomalous grant or login history"

Remediation

Because this campaign exploits trust relationships and identity — not a software flaw — remediation is architectural and procedural, not a patch cycle. Prioritize in this order:

Immediate (24-72 hours if you suspect exposure):

  1. Revoke all third-party OAuth grants with mail/Drive scopes that are not business-critical. In Google Admin Console: Security → API Controls → App Access Control. Restrict third-party API access to an allowlist model — this single control would have broken the persistence mechanism in most COLD RELIC intrusions.
  2. Force credential resets and token revocation for all accounts with anomalous login or grant history. Use gam user <email> deprovision to invalidate sessions, app passwords, and OAuth tokens in one operation.
  3. Audit Gmail forwarding rules and filters. External auto-forwarding is the quietest exfiltration channel in Workspace. Remove any rule forwarding to domains you do not own.
  4. Enforce phishing-resistant MFA. Move high-value users (executives, admins, finance, anyone in partner-facing roles) to hardware security keys or passkeys. TOTP and push-based MFA are defeated by the AiTM phishing kits this actor class favors.
  5. Enable and centralize Google Workspace audit logging into your SIEM. The detections above are worthless if your Workspace logs never leave Google's console.

Strategic (30-90 days):

  1. Adopt a third-party notification playbook. If your tenant is breached, your IR plan must include assessing whose data was exposed and notifying downstream partners. In this campaign, Organization B's defenders never had a chance because Organization A's compromise was their threat intel — and they didn't have it. Mandiant's investigation is the canonical example of why breach scoping must extend to data-about-third-parties.
  2. Implement context-aware access policies. Restrict Workspace access by device posture and geography where operationally feasible, and alert on impossible-travel and impossible-ASN logins.
  3. Hunt your own data as an attack surface. Assume anything in a compromised mailbox becomes a spearphishing template. Brief partner-facing staff that reply-thread hijacking from a "trusted" counterpart is an active technique.
  4. Review the Mandiant/GTIG reporting and map observed TTPs to MITRE ATT&CK techniques T1078.004 (Cloud Accounts), T1550.001 (Web Session Cookie), T1530 (Data from Cloud Storage), and T1566 (Phishing) in your detection coverage matrix.

There is no vendor patch and no CISA deadline here — the deadline is self-imposed. Every day an unreviewed OAuth grant or external forwarding rule persists in your tenant is a day this tradecraft works against you.

Related Resources

Security Arsenal Incident Response Services AlertMonitor Platform Book a SOC Assessment incident-response Intel Hub

Is your security operations ready?

Get a free SOC assessment or see how AlertMonitor cuts through alert noise with automated triage.