Threat reportThreat IntelligenceTL-2026-2881
Desktop AI Supercomputers, Uncensored Models and Agentic Frameworks (HexStrike-AI) Enable Automated Large-Scale Attacks, incl. CVE-2025-7775 Citrix NetScaler
Netlas reports that affordable local AI hardware (NVIDIA DGX Spark), abliterated open-weight models and MCP-based agentic orchestration are converging to make autonomous, large-scale attacks cheap. Check Point documented threat actors pointing the open-source HexStrike-AI MCP framework at Citrix NetScaler CVE-2025-7775 within 12 hours of disclosure, with claims of exploitation compressed from days to under ten minutes. First disclosed .
Key facts
| Identifiers | TL-2026-2881, CVE-2025-7775, CVE-2025-7776, CVE-2025-8424 |
|---|---|
| Also known as | HexStrike-AI abuse, AI-orchestrated automated exploitation, GTG-1002 (referenced context) |
| Severity | HIGH |
| CVSS | 9.8 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H) |
| First disclosed | |
| Exploitation status | Linked CVE exploited in the wild; Listed in CISA KEV: CVE-2025-7775 (added ) |
| EPSS | CVE-2025-7775: 20.2% (97.4th percentile); CVE-2025-7776: 8.2% (94.7th percentile); CVE-2025-8424: 3.3% (88.3th percentile) |
| Affected products | Citrix / Cloud Software Group NetScaler ADC and NetScaler Gateway (CVE-2025-7775) |
| Weaknesses (CWE) | CWE-119 |
| Tools | Abliterated open-weight LLMs (Hugging Face), Claude Code, FastMCP, Heretic, HexStrike-AI, Nmap, Nuclei |
| MITRE ATT&CK / ATLAS techniques | AML.T0018 Manipulate AI Model, AML.T0053 AI Agent Tool Invocation, AML.T0054 LLM Jailbreak, T1046 Network Service Discovery, T1059 Command and Scripting Interpreter, T1078 Valid Accounts, T1110 Brute Force, T1190 Exploit Public-Facing Application, T1505.003 Web Shell, T1588.002 Tool, T1588.006 Vulnerabilities, T1588.007 Artificial Intelligence and 1 more in the attack flow |
| Motivation | UNKNOWN |
| Target sectors | technology, finance, chemical, government administration, telecoms |
| Target regions | Global |
| Detection rules | 9 (1 shown free) |
| Sources | 12 cited sources |
Detection coverage
As of 2026-10-03, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2881 across Splunk SPL, Microsoft KQL and Sigma, covering 10 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: Server-Side Script Requests Under NetScaler Web Paths Indicating Webshell Persistence — Post-Exploitation
NetScaler serves its portal content with .do, .html and .js resources, not .php/.jsp/.aspx. A successful (HTTP 200) request for a server-side script under /vpn/, /logon/, /menu/ or /netscaler/ indicates a dropped webshell. Actors using HexStrike-AI against CVE-2025-7775 claimed webshell deployment for persistence after unauthenticated exploitation.
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1505.003, T1190
- Data sources: citrix_netscaler_web_logs, splunk_cim_web_datamodel, sentinel_commonsecuritylog, webserver_access_logs
- Severity: high
- Confidence: high
Splunk SPL
```Successful requests for server-side script extensions under NetScaler web roots```
| tstats summariesonly=true count min(_time) as firstTime max(_time) as lastTime values(Web.src) as sources from datamodel=Web.Web where Web.status=200 by Web.dest, Web.uri_path
| rename Web.* as *
| regex uri_path="(?i)^/(vpn|logon|menu|netscaler)/.*\.(php|jsp|jspx|aspx)$"
| eval firstTime=strftime(firstTime,"%F %T"), lastTime=strftime(lastTime,"%F %T")
| sort - countMicrosoft KQL
CommonSecurityLog
| where TimeGenerated > ago(24h)
| where RequestURL matches regex @'(?i)/(vpn|logon|menu|netscaler)/[^?]*\.(php|jsp|jspx|aspx)(\?|$)'
| summarize Requests = count(), Sources = make_set(SourceIP, 10), FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated) by DestinationIP, RequestURL
| order by Requests descSigma
title: Webshell-Style Script Request Under NetScaler Web Paths
id: 9d6f3a28-4c1b-4d92-a7e5-1b8c0f2e6a05
status: experimental
description: Successful request for a .php, .jsp, .jspx or .aspx resource under NetScaler web roots, indicating a webshell dropped after CVE-2025-7775 exploitation.
references:
- https://blog.checkpoint.com/executive-insights/hexstrike-ai-when-llms-meet-zero-day-exploitation/
- https://nvd.nist.gov/vuln/detail/cve-2025-7775
author: Threadlinqs Team
date: 2026/10/03
tags:
- attack.persistence
- attack.t1505.003
- attack.initial-access
- attack.t1190
- cve.2025-7775
logsource:
category: webserver
detection:
selection:
cs-uri-stem|re: '(?i)^/(vpn|logon|menu|netscaler)/.*\.(php|jsp|jspx|aspx)$'
sc-status: 200
condition: selection
falsepositives:
- Unusual custom portal themes or customisations that legitimately serve scripts
level: highKnown false positives:
- Custom portal themes or customisations that legitimately serve server-side scripts
Tuning notes: Baseline legitimate paths on your appliances and add exclusions for any custom portal content. The CEF source may not carry an HTTP status; if absent, the KQL runs without it, so review hits manually. Any hit on an appliance exposed before patching warrants file-system forensics for new web-accessible files.
- NetScaler Packet Engine Crash or Restart Consistent with Memory-Overflow Exploitation (CVE-2025-7775) — Exploitation · unlock with TL Creds
- NetScaler Shell Command Execution with Download or Staging Utilities — Post-Exploitation · unlock with TL Creds
- Acquisition of HexStrike-AI, Heretic or Abliterated Open-Weight Models — Tooling Acquisition · unlock with TL Creds
- Offensive Scanner User-Agents Against NetScaler Gateway Paths (Nuclei/Nmap NSE) — Exploitation · unlock with TL Creds
- NetScaler Gateway Authentication Brute-Force Volume from Single Source — Post-Exploitation · unlock with TL Creds
- Offensive Scanner or Cracker Tools Spawned by Python Agent or HexStrike Parent — C2 / Lateral Movement Recon · unlock with TL Creds
- NetScaler Gateway Mass Scan-and-Exploit Burst from Single Source (HexStrike-AI Pattern) — Exploitation · unlock with TL Creds
- HexStrike-AI MCP Server or Client Execution on Endpoint — C2 / Agentic Orchestration · unlock with TL Creds
Indicators of compromise
As of 2026-10-03, Threadlinqs Intelligence tracks 10 indicators of compromise for TL-2026-2881.
Technical analysis
Netlas (Bedrettin Ortak, 2026-08-05) argues that three trends now combine: (1) desktop AI hardware such as the NVIDIA DGX Spark (1 petaFLOP FP4, 128 GB unified memory, 4 TB SSD, $4,699 as of August 2026; fine-tuning of models up to ~70B parameters and inference up to ~200B), (2) abliterated open-weight models, where refusal behaviour is removed by editing weights (technique popularised by FailSpy building on Arditi et al. 2024; automated by the Heretic tool; an Alice research study of April 2026 reported harmful-prompt compliance rising from 5.8% to 98% across 5 models after abliteration), and (3) agentic orchestration over the Model Context Protocol (MCP), which lets a language model drive many offensive tools autonomously.
The concrete incident cited is HexStrike-AI, an MIT-licensed open-source MCP server (GitHub 0x4m4/hexstrike-ai, published by Muhammad Osama / OTT Cybersecurity LLC) that exposes 150+ security tools (Nmap, Nuclei with 4,000+ templates, browser automation, cloud assessment, 12+ specialised agents) to MCP-capable clients such as Claude Desktop, VS Code Copilot, Cursor and Roo Code. Check Point Research (2025-09-02) reported that within about 12 hours of Citrix's 2025-08-26 advisory for three NetScaler zero-days (CVE-2025-7775, CVE-2025-7776, CVE-2025-8424), underground-forum actors were discussing using HexStrike-AI against them. Claims included unauthenticated RCE on CVE-2025-7775, webshell deployment for persistence, parallel scanning of thousands of IPs, and sale of compromised/vulnerable instances. The 'under 10 minutes' timing is an actor claim relayed by Check Point, not independently measured.
CVE-2025-7775 is a memory overflow (CWE-119) in NetScaler ADC/Gateway leading to RCE and/or DoS, CVSS v3.1 9.8 / v4.0 9.2, exploitable without authentication on appliances configured as Gateway (VPN vserver, ICA Proxy, CVPN, RDP Proxy), AAA vservers, LB vservers bound to IPv6 services, or CR vservers of type HDX. It was added to CISA KEV on 2025-08-26 (federal due date 2025-08-28). Shadowserver counted ~28,000 exposed endpoints initially and ~8,000 by 2025-09-02.
For context the article also cites GTG-1002, assessed by Anthropic with high confidence as a Chinese state-sponsored group that in mid-September 2025 used Claude Code with MCP-connected scanners and password crackers to run an espionage campaign against ~30 targets (technology, finance, chemical manufacturing, government), with the AI performing an estimated 80-90% of tactical work and humans making 4-6 decision points per campaign. GTG-1002 bypassed safeguards through role-play (posing as defensive testers) and task decomposition, not abliteration; Claude's hallucinated credentials were a limiting factor. XBOW, a legitimate autonomous pentester that topped the HackerOne US leaderboard in June 2025, is cited as evidence of the speed of autonomous offense. There is no single named adversary or new malware in the Netlas article; this record tracks the capability trend and the HexStrike-AI/NetScaler abuse.
MITRE ATT&CK / ATLAS techniques by tactic:
Persistence
AML.T0018 Manipulate AI Model; T1505.003 Web Shell
Execution
AML.T0053 AI Agent Tool Invocation; T1059 Command and Scripting Interpreter
Defense Evasion
Discovery
T1046 Network Service Discovery
Initial Access
T1078 Valid Accounts; T1190 Exploit Public-Facing Application
Credential Access
Resource Development
T1588.002 Tool; T1588.006 Vulnerabilities; T1588.007 Artificial Intelligence
Reconnaissance
Remediation
Patches
- NetScaler ADC/Gateway 14.1-47.48 or later
- NetScaler ADC/Gateway 13.1-59.22 or later
- NetScaler ADC 13.1-FIPS/NDcPP 13.1-37.241 or later
- NetScaler ADC 12.1-FIPS/NDcPP 12.1-55.330 or later
Immediate actions
- Upgrade NetScaler ADC/Gateway to 14.1-47.48+, 13.1-59.22+, 13.1-FIPS/NDcPP 13.1-37.241+ or 12.1-FIPS/NDcPP 12.1-55.330+ and hunt for webshells on appliances that were exposed before patching
- Inventory NetScaler appliances configured as Gateway, AAA, IPv6 LB or CR/HDX vservers and prioritise them
- Check CISA KEV for CVE-2025-7775 and meet the same-day patch expectation
Workarounds
- Reduce exposure of Gateway/AAA virtual servers to the internet until patched
- Follow vendor mitigations per CISA KEV guidance or discontinue use if mitigations are unavailable
Longer-term hardening
- Build automated same-day patching and validation pipelines for internet-facing edge devices
- Continuous external attack-surface discovery
- Automated detection and response operating at machine speed; behavioural monitoring rather than payload signatures
- Least-privilege scoping and action logging for internal agent/MCP tools
- Threat-intelligence monitoring of underground forums for rapid tooling changes
Affected and fixed versions
- Citrix / Cloud Software Group — NetScaler ADC and NetScaler Gateway (CVE-2025-7775)
Vulnerable versions: 13.1 before 13.1-59.22; 14.1 before 14.1-47.48; 12.1-FIPS/NDcPP before 12.1-55.330; 13.1-FIPS/NDcPP before 13.1-37.241
Fixed in: 13.1-59.22; 14.1-47.48; 12.1-FIPS/NDcPP 12.1-55.330; 13.1-FIPS/NDcPP 13.1-37.241
Timeline
- Bloomberg reports XBOW, an autonomous AI pentester, ranked #1 on the HackerOne US leaderboard (~1,060 submissions), cited by Netlas as evidence of autonomous offense speed
- HexStrike-AI MCP server (150+ tools, MIT licence) released on GitHub in approximately August 2025; ~1,800 stars and 400+ forks within its first month
- Within ~12 hours of the advisory, underground-forum actors discuss pointing HexStrike-AI at the NetScaler flaws, claiming webshell drops, mass scanning and sale of vulnerable instances (per Check Point)
- Citrix discloses and patches CVE-2025-7775, CVE-2025-7776 and CVE-2025-8424 in NetScaler ADC/Gateway; CVE-2025-7775 confirmed exploited in the wild and added to CISA KEV the same day
- CISA KEV federal remediation deadline for CVE-2025-7775
- Check Point Research publishes HexStrike-AI analysis; Shadowserver counts ~8,000 endpoints still vulnerable to CVE-2025-7775, down from ~28,000 the previous week
- Mid-September 2025: Anthropic detects GTG-1002 using Claude Code with MCP-connected tools against ~30 targets, followed by a ten-day investigation and account bans
- Anthropic publishes its report on the first reported AI-orchestrated cyber espionage campaign, assessing with high confidence a Chinese state-sponsored actor
- Alice research (April 2026) reports abliteration raises harmful-prompt compliance from 5.8% to 98% across 5 models and 110 adversarial prompts
- Netlas publishes analysis tying DGX Spark-class hardware, abliterated models and MCP agent frameworks to automated large-scale attacks
Sources cited
- NVD: CVE-2025-7775 NVD (NIST) · Advisory · 2025-08-26
- CISA Known Exploited Vulnerabilities Catalog CISA · Government advisory · 2025-08-26
- Check Point: HexStrike-AI - When LLMs Meet Zero-Day Exploitation Check Point · Research · 2025-09-02
- Anthropic: Disrupting the first reported AI-orchestrated cyber espionage campaign anthropic.com · Vendor report · 2025-11-13
- Wiz Vulnerability Database: CVE-2025-7775 Wiz · Research · 2025-08-26
- Alice abliteration report (April 2026) go.alice.io · Research · 2026-04-01
- HexStrike AI MCP Agents repository GitHub · Code repository · 2025-08-01
- Desktop AI Supercomputers and Automated Attacks (Netlas) netlas.io · Blog · 2026-08-05
- XBOW: Top 1 - how XBOW did it xbow.com · Blog · 2025-06-24
- Hugging Face: Uncensor any LLM with abliteration huggingface.co · Blog · 2024-06-13
- BleepingComputer: Hackers use new HexStrike-AI tool to rapidly exploit n-day flaws BleepingComputer · News · 2025-09-03
- Infosecurity Magazine: Threat Actors Use HexStrike-AI infosecurity-magazine.com · News · 2025-09-04
Other reports on CVE-2025-7775
Previous and next in threat intel
- AI-accelerated intrusions: Microsoft 2026 Digital Defense Report on phishing, public-facing app exploitation, and AI-enabled attacker tooling (s1ngularity, PromptLock, JADEPUFFER/ENCFORGE) Previous report
- Coordinated model-distillation campaign against OpenAI: 15,000+ accounts attempt to extract protected model reasoning, linked to Moonshot AI-associated individuals Next report
Drafted by the Threadlinqs research pipeline from 12 sources; published ; last updated —.