Skip to navigation

Threat reportMalwareTL-2026-2687

The Infostealer Incursion: Stolen Credentials Breach Cloud, Code, and AI Environments

highACTIVE

Wiz Research analysis of NordStellar credential-exposure data shows an industrialized infostealer ecosystem (Lumma, RedLine, Vidar, StealC, Amadey, Miasma, Ghost Loader) accounting for the majority of harvested credentials, with Lumma+RedLine+Vidar alone representing 85.7% of observed theft. Stolen browser sessions, OAuth tokens, and local credential files now routinely grant attackers access to AWS, Azure/Entra ID, GCP, GitHub/GitLab CI/CD, and AI platforms including OpenAI ChatGPT and Anthropic Claude Code.

Key facts

IdentifiersTL-2026-2687, CVE-2023-7028
Also known asInfostealer Incursion Campaign
SeverityHIGH
First disclosed
Exploitation statusLinked CVE exploited in the wild; Listed in CISA KEV: CVE-2023-7028 (added )
EPSSCVE-2023-7028: 94.6% (99.8th percentile)
Affected productsGitLab Inc. GitLab CE/EE (self-hosted); Amazon Web Services AWS CLI / IAM credential and SSO cache files; Microsoft Azure CLI / Entra ID token cache (accessTokens.json, MSAL cache); Google Google Cloud SDK OAuth token stores (credentials.db, access_tokens.db); GitHub Inc. GitHub.com (OAuth tokens, PATs, SSH keys, web sessions); OpenAI ChatGPT web session; Anthropic Claude Code CLI credential store; Red Hat / npm registry @redhat-cloud-services npm packages
Weaknesses (CWE)CWE-640
AttributionInfostealer operators & Initial Access Brokers (low confidence)
MalwareAUDIOFIX, Amadey - S1025, Lumma / LummaC2, Lumma Stealer - S1213, MetaStealer, Miasma, RedLine Stealer - S1240, Stealc, Vidar / Vidar 2.0, miniRAT
ToolsPEASS-ng (linPEAS/winPEAS), nord-stream
MITRE ATT&CK techniquesT1027 Obfuscated Files or Information, T1071.001 Web Protocols, T1078.004 Cloud Accounts, T1082 System Information Discovery, T1195.002 Compromise Software Supply Chain, T1204.002 Malicious File, T1213 Data from Information Repositories, T1497 Virtualization/Sandbox Evasion, T1528 Steal Application Access Token, T1539 Steal Web Session Cookie, T1550.004 Web Session Cookie, T1552.001 Credentials In Files and 4 more in the attack flow
MotivationFINANCIAL
Target sectorstechnology, software development, cryptocurrency, cross-sector enterprise it, government administration
Target regionsGlobal, North America, Europe, india
Detection rules9 (1 shown free)
Sources20 cited sources

Detection coverage

As of 2026-09-27, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2687 across Splunk SPL, Microsoft KQL and Sigma, covering 28 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: Infostealer C2 Beaconing to Known Lumma/RedLine/Vidar/StealC Infrastructure or Abused Legitimate APIs

Detects outbound network traffic to known Lumma/RedLine/Vidar/StealC/Amadey C2 infrastructure -- including community-corroborated sibling domains identified via TL_OSINT_Scan (sha.sm188dvlv.icu, tak.123ful.net, tak.sm188dvlv.icu, tra.12naga.org, rrr.12naga.org) -- as well as abuse of legitimate consumer-platform APIs (Telegram Bot API, Steam, Epic Games) as a C2 transport channel, a technique the community-observed 'Vidar stealer C2 using Telegram, Steam, and Epic Games APIs' campaign employs to blend into normal traffic.

Splunk SPL

index=proxy OR index=network sourcetype IN ("stream:http","pan:threat","zeek:http")
| eval c2_domain=lower(coalesce(dest_domain, url_domain, dest))
| lookup infostealer_c2_iocs.csv domain AS c2_domain OUTPUT malware_family
| eval is_known_c2=if(isnotnull(malware_family), 1, 0)
| eval is_community_sighting=if(c2_domain IN ("sha.sm188dvlv.icu","tak.123ful.net","tak.sm188dvlv.icu","tra.12naga.org","rrr.12naga.org"), 1, 0)
| eval is_abused_legit_c2=if(match(c2_domain,"(?i)(api\.telegram\.org|steamcommunity\.com|epicgames\.com)") AND NOT match(user_agent,"(?i)(SteamClient|EpicGamesLauncher|Telegram Desktop)"), 1, 0)
| where is_known_c2=1 OR is_community_sighting=1 OR is_abused_legit_c2=1
| stats count, values(malware_family) as malware_family, earliest(_time) as first_seen, latest(_time) as last_seen, values(c2_domain) as c2_domain by src_ip, user_agent
| eval avg_interval_sec=round((last_seen-first_seen)/count,1)
| sort -count

Microsoft KQL

let KnownDevSourceIPs = datatable(ip:string) [];
SigninLogs
| where Timestamp > ago(24h)
| where AppDisplayName in ("GitHub","GitLab") or ResourceDisplayName in ("GitHub","GitLab")
| where AuthenticationRequirement == "singleFactorAuthentication"
| where IPAddress !in (KnownDevSourceIPs)
| join kind=inner (
    DeviceNetworkEvents
    | where Timestamp > ago(24h)
    | where RemoteUrl has_any ("api.github.com","gitlab.com")
    | where InitiatingProcessCommandLine has_any ("nord-stream", "glrt-")
) on $left.UserPrincipalName == $right.InitiatingProcessAccountName
| project Timestamp, UserPrincipalName, IPAddress, Location, RemoteUrl, InitiatingProcessCommandLine
| order by Timestamp desc

Sigma

title: Infostealer C2 Beaconing to Known Lumma/RedLine/Vidar/StealC Infrastructure or Abused Legitimate APIs
id: c6d3f1b7-9e24-4a8d-b062-4fbe7c4a8f23
status: experimental
description: Detects outbound network connections to known Lumma/RedLine/Vidar/StealC/Amadey C2 infrastructure -- including community-corroborated sibling infrastructure (sha.sm188dvlv.icu, tak.123ful.net, tak.sm188dvlv.icu, tra.12naga.org, rrr.12naga.org) -- and abuse of legitimate consumer platform APIs (Telegram Bot API, Steam, Epic Games) as a C2 transport/dead-drop channel, per the Wiz "Infostealer Incursion" report and TL_OSINT_Scan community corroboration.
references:
    - https://www.wiz.io/blog/infostealer-incursion-cloud-ai-credentials
author: Threadlinqs Team
date: 2026-09-27
tags:
    - attack.command-and-control
    - attack.t1071.001
    - attack.t1550.004
    - attack.tool.lumma_stealer
    - attack.tool.redline_stealer
    - attack.tool.vidar
logsource:
    category: network_connection
    product: windows
detection:
    selection_known_c2:
        DestinationHostname|contains:
            - 'sha.sm188dvlv.icu'
            - 'tak.123ful.net'
            - 'tak.sm188dvlv.icu'
            - 'tra.12naga.org'
            - 'rrr.12naga.org'
    selection_abused_api:
        DestinationHostname|contains:
            - 'api.telegram.org'
            - 'steamcommunity.com'
            - 'epicgames.com'
        Image|contains:
            - '\Temp\'
            - '\Downloads\'
            - '\AppData\Roaming\'
    filter_known_apps:
        Image|endswith:
            - '\Telegram.exe'
            - '\steam.exe'
            - '\EpicGamesLauncher.exe'
    condition: selection_known_c2 or (selection_abused_api and not filter_known_apps)
falsepositives:
    - Legitimate Telegram Desktop, Steam, or Epic Games Launcher client traffic from their genuine installed binaries
    - Threat-intelligence or sandbox systems intentionally resolving known-bad domains for research
level: critical

Known false positives:

  • Legitimate application traffic to Telegram/Steam/Epic Games APIs from properly signed, known client binaries
  • Threat-intel/sandbox platforms intentionally beaconing to known-bad IOCs for enrichment

Tuning notes: Refresh infostealer_c2_iocs.csv and the hardcoded community-sighting domain list on each TL_OSINT_Scan enrichment cycle; require the abused-legit-API branch to fire only when the initiating process is NOT the vendor's own signed client to avoid alerting on normal Telegram/Steam/Epic usage.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • Trojanized Gaming/Utility Installer Execution Delivering Commodity Infostealer Payload · unlock with TL Creds
  • Cloud CLI and CI/CD Token Exfiltration via Compromised Developer Workstation · unlock with TL Creds
  • Stolen Session-Cookie Replay Against GitHub/GitLab Plus nord-stream CI/CD Secret-Pull Anomaly · unlock with TL Creds
  • Trojanized Installer / Signed-Binary Proxy Execution Preceding Infostealer Drop (Sigma) · unlock with TL Creds
  • Multi-Cloud and AI-Agent Credential Store Harvesting (Sigma) · unlock with TL Creds
  • Infostealer C2 Beaconing to Known Infrastructure or Abused Legitimate APIs (Sigma) · unlock with TL Creds
  • GitLab CVE-2023-7028 Password-Reset Account-Takeover Exploitation Attempt · unlock with TL Creds
  • Multi-Cloud and AI-Agent Credential Store Harvesting (AWS/Azure/GCP/GitHub/Claude Code) · unlock with TL Creds

Indicators of compromise

As of 2026-09-27, Threadlinqs Intelligence tracks 28 indicators of compromise for TL-2026-2687.

28 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

Wiz Research's 'Infostealer Incursion' report (2026-09-25) analyzes credential-exposure telemetry from NordStellar's threat-exposure-management platform to show that commodity infostealer malware has become the primary initial-access vector into cloud, source-control, and AI environments. Three families -- Lumma (LummaC2), RedLine, and Vidar -- account for 85.7% of detected infostealer incidents and are rented on Malware-as-a-Service (MaaS) platforms for as little as a few hundred dollars a month, alongside StealC, Amadey, Miasma, and Ghost Loader. Distribution relies on trojanized gaming installers (fake Roblox.exe, 'Valorant SkinChanger.exe'), abuse of legitimate signed Windows binaries (vbc.exe), and, notably, exploitation of the maximum-severity GitLab authentication-bypass flaw CVE-2023-7028 (CVSS 10.0, CISA KEV) by the Amadey loader to distribute StealC via a compromised self-hosted GitLab instance.

Once resident on a host, these stealers systematically target cloud CLI credential stores rather than just browser passwords: AWS's plaintext ~/.aws/credentials and ~/.aws/sso/cache (OIDC tokens valid up to 90 days); Azure's .azure/accessTokens.json and MSAL token cache, with Vidar 2.0 adding dedicated enumeration of %USERPROFILE%\.azure and the azureProfile.json subscription/tenant manifest; and GCP's credentials.db/access_tokens.db OAuth refresh-token stores and $GOOGLE_APPLICATION_CREDENTIALS service-account keys. Roughly half of tracked infostealer families now actively collect GitHub credentials (OAuth tokens, PATs, SSH keys, and the __Host-user_session_same_site session cookie) and GitLab artifacts (_gitlab_session cookies, glrt--prefixed runner tokens). AI-platform access has joined this target list: OpenAI's __Secure-next-auth.session-token cookie grants full ChatGPT web-session takeover, and Anthropic Claude Code credentials (~/.claude/.credentials.json, macOS Keychain entries, ANTHROPIC_API_KEY) are harvested for resale into an emerging illicit LLM-access market -- corroborated by the fact that privilege-escalation tooling (PEASS-ng's linPEAS/winPEAS) now ships detection rules specifically for ~/.claude credential exposure. Session-cookie theft in particular lets attackers bypass MFA entirely by replaying an already-authenticated session rather than a password.

By platform, stolen-secret volume in the analyzed dataset skews heavily toward AWS (46%), then GCP (13%), GitHub (~10%), and AI platforms (5%, predominantly OpenAI API keys), against a backdrop of 17 billion+ browser cookies stolen annually across the ecosystem and 400+ distinct non-credential secret types harvested. Initial Access Brokers curate and validate stealer logs before reselling verified, working credentials on underground marketplaces -- the gap between infection and resale ranges from hours to months. Downstream abuse of infostealer-sourced credentials has been tied to LAPSUS$ (Azure AD/Entra ID compromise), Scattered Spider (deploys Vidar and Raccoon Stealer alongside helpdesk social-engineering and SIM-swapping per the CISA/FBI joint advisory AA23-320A), and JINX-0164, a financially motivated actor (active since at least mid-2025, techniques overlapping DPRK-linked UNC1069 but not formally attributed) that combines LinkedIn recruiter-lure social engineering with custom macOS malware (the AUDIOFIX Python RAT and MINIRAT Go backdoor) to harvest Keychain, browser, SSH, and 51 different cryptocurrency-wallet-extension credentials, then uses the open-source nord-stream tool against stolen GitHub tokens to pull CI/CD secrets and inject malicious commits into victim repositories under other developers' identities.

The same infection-to-CI/CD-compromise pattern scaled into a supply-chain worm in the Miasma campaign: a Red Hat engineer's GitHub credentials and session cookie, first exposed in stealer logs on 2026-04-13 and again on 2026-05-15, were used on 2026-06-01 to publish 32 trojanized packages under the @redhat-cloud-services npm namespace, propagating via GitHub Actions OIDC-signed builds rather than a stolen npm token -- turning downstream CI pipelines that built from the poisoned packages into an additional propagation vector. Law-enforcement and industry disruption has repeatedly dented but not eliminated the ecosystem: Operation Magnus (2024-10-28) seized RedLine/MetaStealer infrastructure and led to the identification of RedLine's alleged administrator; a Microsoft/FBI-led operation (2025-05-21) seized 2,300+ Lumma C2 domains, yet Lumma had rebuilt infrastructure and resumed spreading (concentrated in India, with US/Europe activity) by February 2026; and a 2026-06 disruption of the Amadey/StealC loader network recovered 27 million stolen credentials, illustrating both the scale of the problem and its resilience to takedown.

MITRE ATT&CK techniques by tactic:

Defense Evasion

T1027 Obfuscated Files or Information; T1497 Virtualization/Sandbox Evasion

Command and Control

T1071.001 Web Protocols

Initial Access

T1078.004 Cloud Accounts; T1195.002 Compromise Software Supply Chain; T1566 Phishing

Discovery

T1082 System Information Discovery

Execution

T1204.002 Malicious File

Collection

T1213 Data from Information Repositories

Credential Access

T1528 Steal Application Access Token; T1539 Steal Web Session Cookie; T1552.001 Credentials In Files; T1555.003 Credentials from Web Browsers

lateral-movement

T1550.004 Web Session Cookie

Resource Development

T1584.001 Domains; T1608.001 Upload Malware

Remediation

Patches

  • GitLab CE/EE 16.1.6, 16.2.9, 16.3.7, 16.4.5, 16.5.6, 16.6.4, 16.7.2 and later (CVE-2023-7028)

Immediate actions

  • Patch self-hosted GitLab CE/EE to 16.1.6/16.2.9/16.3.7/16.4.5/16.5.6/16.6.4/16.7.2 or later to remediate CVE-2023-7028 (CISA KEV, CVSS 10.0), the flaw Amadey has exploited to distribute StealC
  • Rotate and revoke AWS (~/.aws/credentials, ~/.aws/sso/cache), Azure (accessTokens.json, MSAL token cache), and GCP (credentials.db, access_tokens.db, service-account JSON keys) credentials found on any host with a suspected infostealer infection
  • Revoke and rotate GitHub OAuth tokens/PATs/SSH keys and GitLab runner tokens (glrt- prefix) exposed to a compromised developer workstation, and invalidate the corresponding session cookies (__Host-user_session_same_site, _gitlab_session)
  • Rotate ANTHROPIC_API_KEY and OpenAI API keys, and invalidate ~/.claude/.credentials.json and ChatGPT __Secure-next-auth.session-token sessions on any infected host

Workarounds

  • Restrict or disable the self-hosted GitLab password-reset-via-secondary-email flow on unpatched instances
  • Block execution of unsigned or unexpected binaries from user Downloads/Temp directories to disrupt trojanized-installer delivery (fake Roblox.exe, Valorant SkinChanger.exe)

Longer-term hardening

  • Deploy EDR/behavioral detection tuned to infostealer TTPs (bulk credential-file access, browser-data staging, sandbox-evasion API calls) rather than relying on signature AV, since modern stealers execute and exfiltrate within seconds on fully patched, actively-protected systems
  • Enforce phishing-resistant MFA and device-bound/token-bound sessions to blunt session-cookie replay attacks that bypass password-plus-MFA authentication
  • Monitor stealer-log and dark-web marketplaces and subscribe to breach/credential-exposure monitoring (e.g., NordStellar-class platforms) for exposure of corporate credentials
  • Harden CI/CD pipelines with short-lived OIDC-based credentials over long-lived PATs/runner tokens, branch protection, and commit-signing verification to catch pipeline-based propagation as demonstrated by the Miasma npm worm

Affected and fixed versions

  • GitLab Inc. — GitLab CE/EE (self-hosted)
    Vulnerable versions: 16.1.0-16.1.5; 16.2.0-16.2.8; 16.3.0-16.3.6; 16.4.0-16.4.4; 16.5.0-16.5.5; 16.6.0-16.6.3; 16.7.0-16.7.1
    Fixed in: 16.1.6; 16.2.9; 16.3.7; 16.4.5; 16.5.6; 16.6.4; 16.7.2
  • Amazon Web Services — AWS CLI / IAM credential and SSO cache files
    Vulnerable versions: N/A - local credential-file exposure, not a software version
    Fixed in: N/A
  • Microsoft — Azure CLI / Entra ID token cache (accessTokens.json, MSAL cache)
    Vulnerable versions: N/A - local credential-file exposure, not a software version
    Fixed in: N/A
  • Google — Google Cloud SDK OAuth token stores (credentials.db, access_tokens.db)
    Vulnerable versions: N/A - local credential-file exposure, not a software version
    Fixed in: N/A
  • GitHub Inc. — GitHub.com (OAuth tokens, PATs, SSH keys, web sessions)
    Vulnerable versions: N/A - session/token theft, not a software version
    Fixed in: N/A
  • OpenAI — ChatGPT web session
    Vulnerable versions: N/A - session-cookie theft, not a software version
    Fixed in: N/A
  • Anthropic — Claude Code CLI credential store
    Vulnerable versions: N/A - local credential-file/env-var exposure, not a software version
    Fixed in: N/A
  • Red Hat / npm registry — @redhat-cloud-services npm packages
    Vulnerable versions: 32 packages published under the @redhat-cloud-services namespace, 2026-06-01 (Miasma worm)
    Fixed in: Malicious packages removed/unpublished by the npm registry post-disclosure

Timeline

  • StealC infostealer first surfaces in the wild, sold as MaaS for $300/month ($1,000 for six months)
  • GitLab 16.1.0 ships with the access-control flaw later tracked as CVE-2023-7028 (weak password-recovery mechanism)
  • GitLab discloses CVE-2023-7028 (CVSS 10.0), an account-takeover flaw via password reset to an unverified email address
  • CISA adds CVE-2023-7028 to the Known Exploited Vulnerabilities catalog following active exploitation
  • Operation Magnus (US, UK, Belgium, Portugal, Australia) seizes RedLine/MetaStealer C2 servers and domains; alleged RedLine administrator Maxim Rudometov later charged
  • Microsoft- and FBI-led global operation seizes and blocks 2,300+ domains forming Lumma Stealer's C2 backbone
  • JINX-0164 begins targeting cryptocurrency organizations via LinkedIn recruiter-lure social engineering and custom macOS malware (Wiz tracking start, mid-2025)
  • Amadey loader observed exploiting a compromised self-hosted GitLab instance (CVE-2023-7028) to distribute the StealC infostealer
  • Bitdefender confirms Lumma Stealer has rebuilt its command infrastructure post-takedown and resumed spreading, concentrated in India with US/Europe activity
  • A Red Hat engineer's GitHub credentials and active session cookie first appear in infostealer logs (per post-incident Whiteintel review)
  • The same Red Hat engineer's GitHub credentials/session cookie reappear in stealer logs a second time
  • Wiz publishes research on JINX-0164, detailing AUDIOFIX/MINIRAT malware and CI/CD-secret theft via the nord-stream tool
  • Miasma worm publishes 32 trojanized packages under the @redhat-cloud-services npm namespace using the compromised engineer's credentials, propagating via GitHub Actions OIDC-signed builds
  • Coordinated disruption of the Amadey/StealC loader network recovers 27 million stolen credentials; Microsoft publishes technical breakdown of both families
  • Wiz Research publishes 'The Infostealer Incursion,' analyzing NordStellar exposure data across the infostealer-to-cloud/code/AI credential pipeline

Sources cited

Showing 15 of 20 sources, primary sources first.

Other reports on CVE-2023-7028

Previous and next in malware

Drafted by the Threadlinqs research pipeline from 20 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats