Skip to navigation

Threat reportRansomwareTL-2026-2928

Azazel: Gentlemen Ransomware Affiliate Compromises 24+ Organizations via Stolen CI/CD Secrets, Abuses MCP as C2 and Runs LEAKNED Leak Site

highACTIVE

CloudSEK documents Azazel, a Russian-speaking Gentlemen ransomware affiliate who breached more than two dozen organisations in six countries, every one reached through stolen CI/CD secrets or credentials recovered from compromised platforms. He exfiltrated data through a three-hop chain (victim, C2, staging, MEGA), published victims on his own LEAKNED leak site, and used MCP exec_in_session as an operational C2 channel.

Key facts

IdentifiersTL-2026-2928
Also known asThe Gentlemen Files, LEAKNED
SeverityHIGH
Affected productsGitLab GitLab CI/CD (self-hosted, secrets in pipeline variables); Various Cloud/SaaS platforms, PostgreSQL, MinIO, Grafana, Kubernetes, Jasypt-encrypted configs, MCP servers
Weaknesses (CWE)CWE-918, CWE-798, CWE-522, CWE-306
AttributionAzazel (medium confidence)
MalwareGentlemen ransomware, azazel
ToolsPenelope, glato, mega-cmd-server, nord-stream
MITRE ATT&CK techniquesT1021.004 Remote Services: SSH, T1059.004 Command and Scripting Interpreter: Unix Shell, T1059.006 Command and Scripting Interpreter: Python, T1068 Exploitation for Privilege Escalation, T1071.001 Application Layer Protocol: Web Protocols, T1074.001 Data Staged: Local Data Staging, T1074.002 Data Staged: Remote Data Staging, T1078.004 Valid Accounts: Cloud Accounts, T1083 File and Directory Discovery, T1110.002 Brute Force: Password Cracking, T1190 Exploit Public-Facing Application, T1199 Trusted Relationship and 13 more in the attack flow
MotivationFINANCIAL
Target sectorslogistics, insurance, pharmacy, health, medical-devices, artificial-intelligence, government administration, finance, software-saas
Target regionsGlobal (6 countries per CloudSEK), Middle East
Detection rules9 (1 shown free)
Sources7 cited sources

Detection coverage

As of 2026-10-11, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2928 across Splunk SPL, Microsoft KQL and Sigma, covering 36 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: Azazel Credential Decryption and Secret-Sweep Scripts (Jasypt, Grafana Hash Cracking, kubeconfig/SSH Key Sweep)

Detects the post-compromise credential-access scripts Azazel ran inside a compromised AI platform: jasypt_decrypt_all.py (bulk Jasypt decryption with the recovered master key), grafana_crack3.py (offline cracking of Grafana admin hashes) and scan_vectors.py (a ~6.1TB sweep for kubeconfig files, SSH keys and credential-bearing container configs), plus find-based kubeconfig and SSH-key sweeps and the Jasypt decryption CLI.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1552.001, T1552.004, T1110.002, T1059.006
  • Data sources: sysmon_for_linux, microsoft_defender_for_endpoint_linux, linux_process_creation
  • Severity: high
  • Confidence: high

Splunk SPL

index=linux sourcetype="sysmon:linux" EventCode=1
```Azazel credential decrypt and secret sweep scripts```
| eval cmd=lower(CommandLine)
| where match(cmd,"(jasypt_decrypt_all|grafana_crack3|scan_vectors)[.]py") OR match(cmd,"jasyptpbestringdecryptioncli") OR (match(cmd,"find ") AND match(cmd,"kubeconfig") AND match(cmd,"id_rsa|id_ed25519|id_ecdsa"))
| stats count min(_time) as firstTime max(_time) as lastTime values(CommandLine) as commands by host, User, ParentImage
| convert ctime(firstTime) ctime(lastTime)

Microsoft KQL

DeviceProcessEvents
| where Timestamp > ago(24h)
| extend Cmd = tolower(ProcessCommandLine)
| where Cmd matches regex @"(jasypt_decrypt_all|grafana_crack3|scan_vectors)[.]py"
    or Cmd contains "jasyptpbestringdecryptioncli"
    or (Cmd contains "find " and Cmd contains "kubeconfig" and (Cmd contains "id_rsa" or Cmd contains "id_ed25519" or Cmd contains "id_ecdsa"))
| project Timestamp, DeviceName, AccountName, FileName, ProcessCommandLine, InitiatingProcessFileName, InitiatingProcessCommandLine

Sigma

title: Credential Decryption And Secret Sweep Scripts Used By Azazel
id: 91e3b7a0-2c4d-4f58-a6d9-8e0b1f5c3d72
status: experimental
description: Detects jasypt_decrypt_all.py, grafana_crack3.py and scan_vectors.py, the Jasypt decryption CLI, and find-based sweeps for kubeconfig files together with SSH private keys, matching post-compromise credential access by the Azazel Gentlemen affiliate.
references:
  - https://www.cloudsek.com/blog/caught-in-4k-the-gentlemen-files
author: Threadlinqs Team
date: 2026/10/05
tags:
  - attack.credential-access
  - attack.t1552.001
  - attack.t1552.004
  - attack.t1110.002
  - attack.execution
  - attack.t1059.006
logsource:
  category: process_creation
  product: linux
detection:
  selection_scripts:
    CommandLine|contains:
      - 'jasypt_decrypt_all.py'
      - 'grafana_crack3.py'
      - 'scan_vectors.py'
      - 'JasyptPBEStringDecryptionCLI'
  selection_find_a:
    CommandLine|contains|all:
      - 'find '
      - 'kubeconfig'
  selection_find_b:
    CommandLine|contains:
      - 'id_rsa'
      - 'id_ed25519'
      - 'id_ecdsa'
  condition: selection_scripts or (selection_find_a and selection_find_b)
falsepositives:
  - Platform engineers legitimately decrypting Jasypt configuration during maintenance
  - Backup or audit scripts that search for key material
level: high

Known false positives:

  • Platform engineers decrypting Jasypt configuration with the official CLI during maintenance
  • Backup or audit scripts that search the filesystem for key material

Tuning notes: Script-name matches (jasypt_decrypt_all.py, grafana_crack3.py, scan_vectors.py) are near-zero FP and should page. For the generic Jasypt CLI and find sweeps, allow-list approved maintenance hosts and accounts and require execution on servers rather than workstations.

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 and Gold in full, 2 calls a day on a free account)
  • SSRF Against URL-Fetching Endpoint Targeting Internal Services and Metadata · unlock with TL Creds
  • Azazel Ransom Note Delivery Artifacts ([indicator], Scripted /etc/motd Edit) · unlock with TL Creds
  • MCP [indicator] Abuse as C2 and Internet MCP Scanner Fingerprint · unlock with TL Creds
  • GitLab CI/CD Variable Store Enumeration Via API · unlock with TL Creds
  • PostgreSQL Server Spawning Shell and Container-Escape Primitives (nsenter PID 1, unshare Root Mapping) · unlock with TL Creds
  • Bulk Exfiltration Tooling (mega-cmd-server, [indicator], aws s3 sync, pg_dump Pipe, HTTP POST to 9999) · unlock with TL Creds
  • Azazel CI/CD Secrets-Harvesting Tooling Execution (glato, nord-stream, gitlab-watchman, gitleaks, [indicator]) · unlock with TL Creds
  • Network Traffic to Azazel C2, Staging, Leak-Site and Beacon Infrastructure · unlock with TL Creds

Indicators of compromise

As of 2026-10-11, Threadlinqs Intelligence tracks 36 indicators of compromise for TL-2026-2928.

Sample indicators (3 of 36)

  • IP address: 23.236.169[.]183
  • Domain: forgitlab[.]com
  • IP address: 162.220.163[.]26

Defanged, TLP:CLEAR. Full set of 36 indicators: unlock with TL Creds.

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

Technical analysis

CloudSEK's 'Caught in 4K: The Gentlemen Files' (2026-10-05) analyses infrastructure operated by 'Azazel', a Russian-speaking affiliate of the Gentlemen ransomware-as-a-service (RaaS) operation. Azazel used Gentlemen tooling and tradecraft but also ran an independent leak site, LEAKNED, publishing victims and keeping ransom proceeds without sharing revenue with the RaaS operator. CloudSEK reports more than two dozen victims across six countries in logistics, insurance, pharmaceuticals, AI/medical imaging, medical devices and government-adjacent sectors. Active exfiltration was still under way when the infrastructure was observed. Victims were notified under TLP:RED before publication.

Attack chain A (CI/CD secrets harvesting): the operator installed CI/CD and secrets-hunting tooling (glato, nord-stream, gitlab-secrets, gitlab-watchman, gitleaks, brute_odoo.py) and enumerated GitLab CI/CD variable stores and repository history. Recovered secrets included Oracle and PostgreSQL credentials, shipping API credentials and SSH private keys for cloud servers. A single GitLab instance hosted pipelines for two unrelated organisations, so one token compromised both plus three cloud-hosted servers. One SaaS-platform compromise extended to a dozen or more client companies from a single CI/CD token (150+ databases, payment gateways, hundreds of repositories). In a government-linked financial registry, 120,000+ records were deleted after exfiltration. A script, va.py, delivered ransom notes to eight surfaces on six internal hosts: /etc/motd, ATTENTION_SENSITIVE_INFORMATION.txt in /root and /home/ubuntu, the SSH banner via sshd_config, the PostgreSQL cluster_name parameter, the pgAdmin login template, a victim GitLab repository README, and a GitLab Issue opened with a pipeline token.

Attack chain B (deep compromise of an AI platform): initial access was an SSRF in an AI medical-imaging API that fetched user-supplied URLs server-side without validation, reaching internal service discovery, object storage, caching and monitoring. The operator recovered the master key for Jasypt-encrypted cluster configuration (jasypt_decrypt_all.py) and bulk-decrypted database passwords, API keys and service tokens. A hardcoded JWT authentication-bypass token, removed in a later commit but recoverable from git history, gave persistent authenticated access. Grafana admin hashes from an exfiltrated monitoring database were cracked offline (grafana_crack3.py). Object storage was continuously mirrored with the MinIO client (mc mirror, with retry logic), and scan_vectors.py swept about 6.1TB for kubeconfig files, SSH keys and credential-bearing container configs. Other tooling references PostgreSQL COPY TO PROGRAM, UDF/C-extension loading, Redis SSH-key write and an OverlayFS SUID container-escape exploit.

MCP abuse: CloudSEK documents one confirmed instance of the Model Context Protocol exec_in_session tool used as a C2/execution channel in a live criminal campaign. va.py issued exec_in_session calls to an MCP server on 127.0.0.1:35367 with a fixed bearer token; mcp_test.py and recon_mcp.py show iterative development; an 'internet-census-mcp-scanner' fingerprint performed internet-wide MCP port discovery; the MCP client identity 'hermes' was observed. Storage-server output was consistent with an agentic AI assistant answering infrastructure-planning questions.

Infrastructure (50TB+): 23.236.169.183 (open directory on 8000, upload listener on 9999, Penelope reverse-shell handler, MCP C2 tooling); 162.220.163.26 (hostname forgitlab / forgitlab.com, GitLab masquerade staging, ~6TB active across 24+ victim directories); 66.179.30.155 ('novostnik', LEAKNED frontend, evidence-badge store and ~22TB vault); 66.203.124.135:443 (MEGA, final destination through mega-cmd-server); 141.95.252.30 (beacon check-in). Exfiltration used aws s3 sync, scp, pg_dump, HTTP POST to port 9999 and custom per-victim dump scripts, consistent with the Gentlemen affiliate shift from Rclone to Restic to mc.

Context: Gentlemen is a human-operated RaaS that emerged in mid-2025, moved to an affiliate model in September 2025, advertises a 90% affiliate share, and uses a Go-based locker; its backend database and chats were leaked in May 2026. Initial access in the wider operation relies heavily on edge-device and stolen-credential access. No CVEs are cited in the CloudSEK report.

MITRE ATT&CK techniques by tactic:

Lateral Movement

T1021.004 Remote Services: SSH

Execution

T1059.004 Command and Scripting Interpreter: Unix Shell; T1059.006 Command and Scripting Interpreter: Python

Privilege Escalation

T1068 Exploitation for Privilege Escalation; T1611 Escape to Host

Command and Control

T1071.001 Application Layer Protocol: Web Protocols

Collection

T1074.001 Data Staged: Local Data Staging; T1074.002 Data Staged: Remote Data Staging; T1213.003 Data from Information Repositories: Code Repositories; T1530 Data from Cloud Storage

Initial Access

T1078.004 Valid Accounts: Cloud Accounts; T1190 Exploit Public-Facing Application; T1199 Trusted Relationship

Discovery

T1083 File and Directory Discovery; T1613 Container and Resource Discovery

Credential Access

T1110.002 Brute Force: Password Cracking; T1552.001 Unsecured Credentials: Credentials In Files; T1552.004 Unsecured Credentials: Private Keys

Impact

T1485 Data Destruction; T1491.001 Defacement: Internal Defacement

Exfiltration

T1567.002 Exfiltration Over Web Service: Exfiltration to Cloud Storage

Resource Development

T1583.001 Acquire Infrastructure: Domains; T1583.003 Virtual Private Server; T1588.002 Obtain Capabilities: Tool

Reconnaissance

T1595.002 Active Scanning: Vulnerability Scanning

Remediation

Immediate actions

  • Block and hunt for 23.236.169.183, 162.220.163.26, 66.179.30.155, 141.95.252.30 and forgitlab.com in network and proxy logs
  • Rotate all CI/CD variables, runner and pipeline tokens, SSH keys and cloud credentials reachable from GitLab; audit CI/CD variable reads from unexpected IPs
  • Alert on MCP initialize RPCs from unapproved client identities such as 'hermes' and on the 'internet-census-mcp-scanner' fingerprint
  • Review README or Issue modifications made by pipeline tokens and ransom-note artifacts (ATTENTION_SENSITIVE_INFORMATION.txt, altered /etc/motd, SSH banner, PostgreSQL cluster_name)

Workarounds

  • Restrict egress to MEGA and unapproved object-storage endpoints from servers and CI runners

Longer-term hardening

  • Move secrets from raw pipeline variables to a dedicated secrets manager; mask and protect every variable and rotate on a fixed cadence
  • Scan git history for committed secrets and treat removed commits as exposed
  • Bind MCP servers strictly to loopback and treat exec_in_session as a privileged, audited operation
  • Do not treat Jasypt ENC() values as a security boundary; keep keys and kubeconfig in a secrets manager
  • Restrict MinIO buckets to least-privilege service accounts and monitor mc mirror / bulk sync from unexpected sources
  • Keep backup credentials on separate infrastructure and test restores; restrict PostgreSQL COPY TO PROGRAM and C-extension loading
  • Validate SSRF-prone URL-fetching endpoints with allow-lists and block access to internal metadata and service endpoints

Affected and fixed versions

  • GitLab — GitLab CI/CD (self-hosted, secrets in pipeline variables)
    Vulnerable versions: Misconfiguration, not a version-specific flaw
  • Various — Cloud/SaaS platforms, PostgreSQL, MinIO, Grafana, Kubernetes, Jasypt-encrypted configs, MCP servers
    Vulnerable versions: Misconfiguration and credential exposure

Timeline

Last updated · 2 updates · revalidated 2× · latest source

  • Gentlemen RaaS emerges (mid-2025) as a human-operated, Go-locker-based ransomware-as-a-service operation
  • Gentlemen operators show activity about a month before the official RaaS launch (approximate, per Unit 42 via secondary sources).
  • Gentlemen moves to an affiliate model advertising a 90% affiliate share of ransoms
  • Gentlemen internal backend database and chats leaked after the administrator reports a hosting-provider compromise
  • Gentlemen record peak monthly victim claims (117 in June 2026, per Unit 42 via secondary sources).
  • Gentlemen's claimed victim total reaches about 580 across 77 countries by one tracker's count
  • CloudSEK publishes 'Caught in 4K: The Gentlemen Files' after TLP:RED notification of named victims
  • CloudSEK observes Azazel's C2, staging and LEAKNED infrastructure with exfiltration to MEGA still active; 24+ victims across six countries identified
  • Cyber Security News reports the CloudSEK findings and lists the actor's C2, staging and leak-site infrastructure.

Update history

Sources cited

More from Azazel

Previous and next in ransomware

Drafted by the Threadlinqs research pipeline from 7 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