Skip to navigation

Threat reportRansomwareTL-2026-4921

Aur0ra Ransomware Operations Use Cursor and Claude AI to Automate Attacks (GOLD OGRE)

highACTIVE

GOLD OGRE, a financially motivated and assessed Russian-speaking cybercrime group, operates the closed-program Aur0ra ransomware and used the Cursor AI coding agent (backed by Anthropic Claude Sonnet) to plan and run Active Directory discovery, credential attacks, NTLM relay, AD CS abuse and S3-compatible exfiltration. First disclosed .

Key facts

IdentifiersTL-2026-4921
Also known asAur0ra ransomware, Aurora ransomware
SeverityHIGH
First disclosed
Affected productsMicrosoft Windows Active Directory / AD CS environments; VMware ESXi / vCenter; Veeam and other backup platforms Backup servers and NAS
AttributionGOLD OGRE (medium confidence)
MalwareAur0ra, Aurora Ransomware, Cobalt Strike
ToolsBloodHound - S0521, Certipy, Cobalt Strike, ISL Light, Impacket - S0357, PetitPotam, Xray-core, s5cmd
MITRE ATT&CK techniquesT1003 OS Credential Dumping, T1018 Remote System Discovery, T1021.001 Remote Services: Remote Desktop Protocol, T1021.002 Remote Services: SMB/Windows Admin Shares, T1021.006 Remote Services: Windows Remote Management, T1046 Network Service Discovery, T1053.005 Scheduled Task/Job: Scheduled Task, T1055.012 Process Injection: Process Hollowing, T1059.001 Command and Scripting Interpreter: PowerShell, T1069.002 Permission Groups Discovery: Domain Groups, T1070.006 Indicator Removal: Timestomp, T1071.001 Application Layer Protocol: Web Protocols and 27 more in the attack flow
MotivationFINANCIAL
Target sectorsmanufacturing, food-and-agriculture, pharmaceuticals, professional-services, title-and-legal
Target regionsNorth America, Europe
Detection rules9 (1 shown free)
Sources8 cited sources

Detection coverage

As of 2026-11-04, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-4921 across Splunk SPL, Microsoft KQL and Sigma, covering 46 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: Aur0ra GOLD OGRE AD Attack Tooling - Certipy, NTLM Relay/Coercion, NetExec, BloodHound, GodPotato

Behaviour-first detection of the Active Directory attack tooling GOLD OGRE operators (and their Cursor AI agent) run against victim domains for the Aur0ra ransomware operation: Certipy AD CS abuse (ESC1/ESC6/ESC8), Impacket ntlmrelayx with PetitPotam/PrinterBug/Coercer coercion, NetExec and SharpShares enumeration, SharpHound/BloodHound collection, nltest domain discovery and GodPotato token impersonation. Hosts that launch several distinct tools are ranked first.

Splunk SPL

index=sysmon sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational" EventCode=1 earliest=-24h
    (Image="*GodPotato*" OR Image="*SharpShares*" OR Image="*SharpHound*" OR Image="*PetitPotam*" OR Image="*Certipy*"
     OR CommandLine="*ntlmrelayx*" OR CommandLine="*certipy find*" OR CommandLine="*certipy req*" OR CommandLine="*certipy auth*" OR CommandLine="*certipy relay*" OR CommandLine="*certipy-ad*"
     OR CommandLine="*petitpotam*" OR CommandLine="*printerbug*" OR CommandLine="*coercer*"
     OR CommandLine="*nxc smb*" OR CommandLine="*nxc ldap*" OR CommandLine="*nxc winrm*" OR CommandLine="*netexec*" OR CommandLine="*bloodhound-python*"
     OR (CommandLine="*nltest*" AND (CommandLine="*/dclist*" OR CommandLine="*/domain_trusts*")))
| eval cmdblob=lower(coalesce(Image,"")." ".coalesce(CommandLine,""))
| eval tool=case(match(cmdblob,"godpotato"),"GodPotato token impersonation (T1134.001)", match(cmdblob,"certipy"),"Certipy AD CS abuse (T1649)", match(cmdblob,"ntlmrelayx|petitpotam|printerbug|coercer"),"NTLM coercion or relay (T1557.001)", match(cmdblob,"sharpshares"),"SharpShares share discovery (T1135)", match(cmdblob,"sharphound|bloodhound"),"BloodHound collection (T1069.002)", match(cmdblob,"nxc |netexec"),"NetExec AD enumeration (T1087.002)", match(cmdblob,"nltest"),"nltest domain discovery (T1087.002)", true(),"other AD tooling")
| stats count min(_time) as first_seen max(_time) as last_seen dc(tool) as distinct_tools values(tool) as tools values(CommandLine) as commands by host, User
| eval first_seen=strftime(first_seen,"%F %T"), last_seen=strftime(last_seen,"%F %T")
| sort 0 -distinct_tools -count

Microsoft KQL

DeviceProcessEvents
| where Timestamp > ago(24h)
| where FileName has_any ("GodPotato", "SharpShares", "SharpHound", "PetitPotam", "Certipy")
    or ProcessCommandLine has_any ("ntlmrelayx", "certipy find", "certipy req", "certipy auth", "certipy relay", "certipy-ad", "petitpotam", "printerbug", "coercer", "nxc smb", "nxc ldap", "nxc winrm", "netexec", "bloodhound-python")
    or (ProcessCommandLine has "nltest" and ProcessCommandLine has_any ("/dclist", "/domain_trusts"))
| extend Blob = tolower(strcat(FileName, " ", ProcessCommandLine))
| extend Tool = case(
    Blob has "godpotato", "GodPotato token impersonation (T1134.001)",
    Blob has "certipy", "Certipy AD CS abuse (T1649)",
    Blob has_any ("ntlmrelayx", "petitpotam", "printerbug", "coercer"), "NTLM coercion or relay (T1557.001)",
    Blob has "sharpshares", "SharpShares share discovery (T1135)",
    Blob has_any ("sharphound", "bloodhound"), "BloodHound collection (T1069.002)",
    Blob has_any ("nxc ", "netexec"), "NetExec AD enumeration (T1087.002)",
    Blob has "nltest", "nltest domain discovery (T1087.002)",
    "other AD tooling")
| summarize Events = count(), DistinctTools = dcount(Tool), Tools = make_set(Tool), Commands = make_set(ProcessCommandLine, 10), FirstSeen = min(Timestamp), LastSeen = max(Timestamp) by DeviceName, AccountName
| order by DistinctTools desc, Events desc

Sigma

title: Aur0ra GOLD OGRE Active Directory Attack Tooling Execution
id: 3f6a1c52-8d47-4e0b-9a1e-5b7c2d9e4a11
status: experimental
description: Detects execution of the AD discovery, NTLM coercion/relay, AD CS abuse and token impersonation tooling used by GOLD OGRE (Aur0ra ransomware) - Certipy, Impacket ntlmrelayx, PetitPotam, PrinterBug, Coercer, NetExec, SharpShares, SharpHound/BloodHound and GodPotato.
references:
    - https://www.sophos.com/en-us/threat-profiles/gold-ogre
    - https://activesoc.blackhillsinfosec.com/blog/introducing-the-aur0ra-ransomware-group/
    - https://gbhackers.com/aur0ra-ransomware-ai-attacks/
author: Threadlinqs Team
date: 2026-10-31
tags:
    - attack.credential_access
    - attack.t1649
    - attack.t1557.001
    - attack.discovery
    - attack.t1087.002
    - attack.t1069.002
    - attack.t1135
    - attack.privilege_escalation
    - attack.t1134.001
    - attack.s0357
    - attack.s0521
logsource:
    category: process_creation
    product: windows
detection:
    selection_img:
        Image|contains:
            - 'GodPotato'
            - 'SharpShares'
            - 'SharpHound'
            - 'PetitPotam'
            - 'Certipy'
    selection_cli:
        CommandLine|contains:
            - 'ntlmrelayx'
            - 'certipy find'
            - 'certipy req'
            - 'certipy auth'
            - 'certipy relay'
            - 'certipy-ad'
            - 'petitpotam'
            - 'printerbug'
            - 'coercer'
            - 'nxc smb'
            - 'nxc ldap'
            - 'nxc winrm'
            - 'netexec'
            - 'bloodhound-python'
    selection_nltest:
        Image|endswith: '\nltest.exe'
        CommandLine|contains:
            - '/dclist'
            - '/domain_trusts'
    condition: 1 of selection_*
falsepositives:
    - Authorized red team or penetration test using Impacket, NetExec, Certipy or BloodHound
    - Administrators running nltest /dclist or /domain_trusts during troubleshooting
level: high

Known false positives:

  • Authorized red team or penetration test using Impacket, NetExec, Certipy or BloodHound
  • Administrators running nltest /dclist or /domain_trusts during troubleshooting

Tuning notes: Allowlist approved assessment hosts and service accounts by host/User. Raise the signal by alerting when distinct_tools >= 2 on one host within 24h. Tools are often renamed; pair with AD CS certificate-request audit events (4886/4887) and EDR command-line telemetry. Linux-hosted operator tooling (Certipy/ntlmrelayx run from a SOCKS-proxied host) will not appear in Windows process telemetry - monitor for the resulting 4768/4769/4887 anomalies instead.

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)
  • Aur0ra GOLD OGRE Password Spraying and Valid-Account Logons From Operator Infrastructure · unlock with TL Creds
  • Aur0ra Data Staging and Exfiltration - PowerShell 7-Zip Split Archives and s5cmd S3 Copy · unlock with TL Creds
  • Aur0ra GOLD OGRE Remote Access Persistence - ISL Light, Quick Assist and Scheduled-Task Xray-core Masquerade Binaries · unlock with TL Creds
  • Aur0ra GOLD OGRE Kerberos Credential Attacks - Kerberoasting RC4 Ticket Bursts and PKINIT Certificate TGT Requests · unlock with TL Creds
  • Aur0ra ESXi Force-Kill of VMs, [indicator] Execution and [indicator] Discovery · unlock with TL Creds
  • Aur0ra GOLD OGRE Pass-the-Hash and Admin Share Lateral Movement - seclogo Type 9, NTLM Type 3 and ADMIN$ Binary Writes · unlock with TL Creds
  • Aur0ra Ransomware Inhibit System Recovery - Shadow Copy Deletion, SystemRestore Disable, Event Log Clearing, [indicator] and Ransom Note · unlock with TL Creds
  • Aur0ra GOLD OGRE Xray-core Proxy C2 - [indicator]/[indicator] and Operator Infrastructure · unlock with TL Creds

Indicators of compromise

As of 2026-11-04, Threadlinqs Intelligence tracks 46 indicators of compromise for TL-2026-4921.

Sample indicators (3 of 46)

  • IP address: 152.236.4[.]14
  • Domain: ijexszhscln27nl263lmcd7tx3jttkhm4wjhd4e3y6r4csdbfyeprvid[.]onion
  • File hash: 81ca5fc6b55accdbc44266d66bd72c7c4152a75b215593adc433d51250054333

Defanged, TLP:CLEAR. Full set of 46 indicators: unlock with TL Creds.

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

Technical analysis

Sophos CTU, CloudSEK and Gambit Security document a multi-victim operation spanning 20+ organizations in nine countries, with human operators supervising the AI agent.

Aur0ra (also written Aurora) ransomware is operated by the group Sophos tracks as GOLD OGRE. Sophos describes it as a closed program rather than an openly advertised RaaS. It has been active since roughly April 2026, with about 40 published victims through September 2026 according to Sophos. Victims are concentrated in the United States and Western Europe, dwell time ranges from 12 hours to 9 days (about 12 days from intrusion to leak-site post per Black Hills InfoSec), and ransom demands have exceeded USD 1 million.

Initial access is reported in two forms. Sophos documents the use of stolen credentials against internet-exposed VPN services. Black Hills InfoSec and the CloudSEK-based reporting describe email-bombing followed by vishing calls that impersonate IT help desk staff, with Xray-core (VLESS/REALITY, masqueraded as ChromeUpdate.exe and ConnectivityHost.exe) used for command and control and proxying. Persistence and lateral movement use Cobalt Strike Beacon, ISL Light, pass-the-hash, RDP, WinRM, SMB, LDAP and SOCKS5 proxying. Reconnaissance uses SharpShares, nltest, Nmap, NetExec and BloodHound. Privilege escalation and lateral movement include NTLM coercion and relay (PetitPotam, Coerce Plus, PrinterBug, Impacket ntlmrelayx), Kerberoasting, and Active Directory Certificate Services abuse with Certipy (ESC1, ESC6 and ESC8 referenced). GodPotato was also observed.

AI use is the distinguishing feature. Gambit Security recovered 28 Cursor chat sessions covering ten target organizations between April 8 and May 21, 2026. CloudSEK analyzed an exposed open directory with shell history covering April to July 2026 and 20+ organizations in nine countries. Operators wrote instructions to the Cursor agent in Russian, excluded CIS ranges and domains, and forbade noisy actions such as DCSync, account lockouts and brute-forcing. Most AI-issued commands failed on the first attempt and needed manual refinement, so the activity was human-supervised rather than autonomous. Per the GBHackers summary of Sophos CTU research, operators first relied almost entirely on Anthropic models, then added xAI and Moonshot models. Humans supplied VPN credentials, SOCKS access, targets and objectives, and the agents ran AD discovery, analyzed credential attack paths and moved laterally, prioritizing backup servers, NAS and virtualization hosts.

Data is staged with PowerShell-driven 7-Zip archiving (reported in 50 GB chunks) and exfiltrated to S3-compatible storage with s5cmd (a VeeamTransport-named binary was also used for S3 copying). SQL Server xp_cmdshell abuse was reported. The Windows locker (sap.exe in CloudSEK reporting) deletes shadow copies, disables System Restore and encrypts files in place without changing extensions, dropping !!!README!!!DO_NOT_DELETE.txt. The Linux/ESXi variant (encrypt.out) force-kills VMs with esxcli and encrypts VM disk and state files. Reporting on the cryptography differs between sources: CloudSEK-based reporting cites ChaCha20 with RSA-4096 key wrapping and a Zig codebase, while Cohesity cites the Windows CryptoAPI. Treat the algorithm details as unreconciled.

No CVEs are associated with this threat. Victim counts differ by source and date (3 in the GBHackers article, 4 leak-site victims in THN/CloudSEK, 28 per Ransomware.live via Black Hills, about 40 per Sophos). Severity HIGH is an analyst assignment. BeaconBeagle lookups for the three article IPs returned HTTP 404 (no C2 record).

MITRE ATT&CK techniques by tactic:

Credential Access

T1003 OS Credential Dumping; T1110.003 Brute Force: Password Spraying; T1187 Forced Authentication; T1552.001 Unsecured Credentials: Credentials In Files; T1555.003 Credentials from Password Stores: Credentials from Web Browsers; T1557.001 Name Resolution Poisoning and SMB Relay; T1558.003 Steal or Forge Kerberos Tickets: Kerberoasting; T1558.004 Steal or Forge Kerberos Tickets: AS-REP Roasting; T1649 Steal or Forge Authentication Certificates

Discovery

T1018 Remote System Discovery; T1046 Network Service Discovery; T1069.002 Permission Groups Discovery: Domain Groups; T1087.002 Account Discovery: Domain Account; T1135 Network Share Discovery; T1482 Domain Trust Discovery

Lateral Movement

T1021.001 Remote Services: Remote Desktop Protocol; T1021.002 Remote Services: SMB/Windows Admin Shares; T1021.006 Remote Services: Windows Remote Management; T1550.002 Use Alternate Authentication Material: Pass the Hash

Persistence

T1053.005 Scheduled Task/Job: Scheduled Task

Defense Evasion

T1055.012 Process Injection: Process Hollowing; T1070.006 Indicator Removal: Timestomp

Execution

T1059.001 Command and Scripting Interpreter: PowerShell

Command and Control

T1071.001 Application Layer Protocol: Web Protocols; T1090 Proxy; T1219 Remote Access Tools; T1572 Protocol Tunneling

Initial Access

T1078 Valid Accounts; T1133 External Remote Services; T1566.004 Phishing: Spearphishing Voice

Privilege Escalation

T1134.001 Access Token Manipulation: Token Impersonation/Theft

Impact

T1486 Data Encrypted for Impact; T1490 Inhibit System Recovery

Collection

T1560.001 Archive Collected Data: Archive via Utility

Exfiltration

T1567 Exfiltration Over Web Service

Resource Development

T1587.001 Develop Capabilities: Malware; T1588.002 Obtain Capabilities: Tool

defense-impairment

T1685 Disable or Modify Tools; T1685.005 Clear Windows Event Logs

Remediation

Patches

  • No CVEs are associated with this threat; patch internet-facing devices as general hygiene

Immediate actions

  • Block and hunt for 152.236.4.14, 45.59.113.63, 172.86.113.245, 172.86.116.200 and 151.244.30.43 in network telemetry
  • Hunt for sap.exe, encrypt.out, ChromeUpdate.exe, ConnectivityHost.exe and !!!README!!!DO_NOT_DELETE.txt
  • Alert on vssadmin delete shadows, wmic shadowcopy delete and schtasks disabling the SystemRestore task in quick succession
  • Audit outbound connections for tunneling tools (VPN clients, proxychains, Xray-core) and s5cmd or S3-compatible uploads
  • Preserve AI coding-assistant session logs and API request history if unauthorized use is found

Workarounds

  • Rotate credentials exposed to VPN and help-desk social engineering
  • Rotate the krbtgt password twice after domain compromise

Longer-term hardening

  • Deploy phishing-resistant MFA on VPN and remote access
  • Segment critical servers and isolate backups and ESXi/vCenter on separate credentials and networks
  • Review domain admin privileges and implement privileged access management
  • Remove SPNs from privileged accounts and migrate to group-managed service accounts
  • Disable LLMNR and NBT-NS, enforce SMB signing, restrict WinRM
  • Audit AD CS templates for ESC1, ESC6 and ESC8 misconfigurations
  • Monitor EDR alerts with rapid response and deploy application controls
  • Treat commercial coding agents as privileged software and scope their credentials to the minimum

Affected and fixed versions

  • Microsoft — Windows Active Directory / AD CS environments
    Vulnerable versions: Misconfigured AD CS templates (ESC1, ESC6, ESC8); Domains without SMB signing or with NTLM relay exposure
  • VMware — ESXi / vCenter
    Vulnerable versions: Hosts reachable from compromised enterprise networks
  • Veeam and other backup platforms — Backup servers and NAS
    Vulnerable versions: Backups not isolated from domain credentials

Timeline

Last updated · 3 updates · revalidated 3× · latest source

  • Start of the window (April 8 to May 21, 2026) covered by 28 recovered Cursor agent chat sessions across ten victim organizations, per Gambit Security.
  • Aur0ra ransomware group first reported at the end of April 2026 (Black Hills InfoSec); Sophos says GOLD OGRE has been active since April 2026.
  • End of the Gambit-observed window of Cursor-assisted exploitation against ten organizations.
  • Sophos CTU identified the AI-assisted post-exploitation framework in May 2026 (month per GBHackers); a CYFIRMA weekly report dated 2026-05-22 also appears in search results for Aur0ra and GOLD OGRE.
  • Sophos publishes research on an AI-assisted malware development lab (Claude Opus 4.5 coordinating agents via MCP, ~80 modules, 70+ evasion techniques tested against Sophos, CrowdStrike and Microsoft Defender).
  • CloudSEK's analysis of an exposed open directory (shell history) covers April to July 2026 activity against 20+ organizations in nine countries.
  • Sophos CTU publishes 'AI used in Aur0ra ransomware testing and execution' (exact day unconfirmed; August 2026), attributing live Cursor-assisted intrusions to GOLD OGRE and listing Cobalt Strike, exfiltration and SOCKS infrastructure.
  • Gambit Security report and Reuters coverage disclose Aur0ra operators' use of Cursor's AI agent against seven companies in six countries.
  • CloudSEK TRIAD publishes 'Caught in 4K: The Aurora Files' with locker hashes, operator VPS and SOCKS pivot IPs, onion negotiation site and payment analysis.
  • CloudSEK and Gambit Security findings on infrastructure exposure and Cursor use published, covered by The Hacker News in late August 2026.
  • Sophos GOLD OGRE threat profile reports nearly 40 victims published on the Aur0ra leak site through September 2026.
  • Cloud Security Alliance published a research note on the Aurora ransomware abuse of Cursor AI.
  • UVCyber threat advisory on Aurora ransomware and AI-assisted exploitation is issued.
  • Sophos CTU publishes 'AI used in Aur0ra ransomware testing and execution', linking the AI-orchestrated EDR testing framework to the C2 and SOCKS infrastructure (including locked-ioc-24) shared with Aur0ra ransomware operations; Sophos assesses with high confidence that AI was used in both testing and execution.
  • GBHackers reported that GOLD OGRE used Cursor and Claude to automate Aur0ra operations, with three victims on the leak site and IOCs listed.
  • Help Net Security, citing Sophos, reports three victims that reached domain administrator compromise were named on the Aur0ra data leak site.

Update history

Sources cited

Previous and next in ransomware

Drafted by the Threadlinqs research pipeline from 8 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