Skip to navigation

Threat reportVulnerabilityTL-2026-2358

StyleSmuggler — Unpatched Magento and Adobe Commerce Zero-Day Exploited to Backdoor Online Stores

criticalACTIVE

An unauthenticated remote code execution zero-day vulnerability in Magento Open Source and Adobe Commerce, discovered by Sansec and named StyleSmuggler. The two-stage attack chain abuses Magento's GraphQL endpoint and template rendering system to inject PHP code into log files, then triggers execution through the built-in Payment Transaction Failed Reminder email template — deploying a persistent Rust backdoor disguised as a Linux kernel thread.

Key facts

IdentifiersTL-2026-2358, CVE-2026-75650
Also known asStyleSmuggler
SeverityCRITICAL
CVSS10
First disclosed
Exploitation statusLinked CVE exploited in the wild; Listed in CISA KEV: CVE-2026-75650 (added )
EPSSCVE-2026-75650: 3.9% (90.1th percentile)
Affected productsAdobe Adobe Commerce; Magento Magento Open Source
Weaknesses (CWE)CWE-94, CWE-74, CWE-73, CWE-502, CWE-1336
Malwaregvfsd-user
MITRE ATT&CK techniquesT1001.003 Protocol or Service Impersonation, T1005 Data from Local System, T1008 Fallback Channels, T1027 Obfuscated Files or Information, T1036.004 Masquerading: Masquerade Task or Service, T1036.005 Match Legitimate Resource Name or Location, T1041 Exfiltration Over C2 Channel, T1048.003 Exfiltration Over Unencrypted Non-C2 Protocol, T1053.003 Scheduled Task/Job: Cron, T1059 Command and Scripting Interpreter, T1059.004 Command and Scripting Interpreter: Unix Shell, T1059.006 Command and Scripting Interpreter: Python and 24 more in the attack flow
MotivationFINANCIAL
Target sectorsecommerce, retail, online-store
Target regionsGlobal
Detection rules9 (1 shown free)
Sources11 cited sources

Detection coverage

As of 2026-11-07, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2358 across Splunk SPL, Microsoft KQL and Sigma, covering 82 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: StyleSmuggler C2 Communication — WebSocket TLS and NTP-Shaped Beaconing

Detects the StyleSmuggler Rust backdoor's dual-channel C2 infrastructure: WebSocket over TLS (port 443) to 99.84.67.186 and windwsecurity.run for remote shell access, and custom NTP-shaped UDP traffic (port 123) to typosquatted domains ntp.timesysnc.net, time.microsft.run, and pool.microsft.studio blending with legitimate NTP traffic. Also detects malware download from 247.cdnflare.xyz.

Splunk SPL

index=network sourcetype=pan_log
| search (dest_ip=99.84.67.186 OR dest_domain=windwsecurity.run OR dest_domain=247.cdnflare.xyz OR dest_domain=ntp.timesysnc.net OR dest_domain=time.microsft.run OR dest_domain=pool.microsft.studio)
| eval c2_type=case(
    dest_port=443 AND (like(dest_domain,"%windwsecurity%") OR dest_ip="99.84.67.186"), "WebSocket TLS C2",
    dest_port=123 AND (like(dest_domain,"%timesysnc%") OR like(dest_domain,"%microsft%")), "NTP-shaped C2",
    like(dest_domain,"%cdnflare%"), "Malware Download",
    1=1, "C2 Infrastructure")
| eval attack_stage="StyleSmuggler - C2"
| table _time, src_ip, dest_ip, dest_domain, dest_port, protocol, action, bytes, app, c2_type
| sort - _time

Microsoft KQL

DeviceNetworkEvents
| where Timestamp > ago(24h)
| where (RemotePort == 443 and (RemoteUrl has_any ("windwsecurity.run", "247.cdnflare.xyz") or RemoteIP == "99.84.67.186"))
    or (RemotePort == 123 and Protocol in ("Udp", "UDP") and RemoteUrl has_any ("ntp.timesysnc.net", "time.microsft.run", "pool.microsft.studio"))
| extend C2Type = case(
    RemotePort == 123, "NTP-shaped C2 beacon",
    RemoteUrl has "windwsecurity.run", "WebSocket TLS C2 (remote shell)",
    RemoteUrl has "cdnflare.xyz", "Malware download host",
    RemoteIP == "99.84.67.186", "WebSocket TLS C2",
    "Suspicious C2 connection"
)
| project Timestamp, DeviceName, RemoteUrl, RemoteIP, RemotePort, Protocol, LocalIP, LocalPort, ActionType, C2Type
| sort by Timestamp desc

Sigma

title: StyleSmuggler C2 Communication Over NTP-Shaped and WebSocket TLS Channels
id: 1b4f7a2c-8d3e-4f5a-9b6c-2d7e3f8a1b4c
status: experimental
description: Detects the StyleSmuggler Rust backdoor's C2 communication patterns including WebSocket over TLS to known C2 servers and custom NTP-shaped UDP traffic to typosquatted NTP domains used for stealthy beaconing.
references:
  - https://sansec.io/research/stylesmuggler
  - https://github.com/disrex-group/stylesmuggler-mitigation/blob/main/IOC.md
  - https://github.com/disrex-group/stylesmuggler-mitigation
author: Threadlinqs Team
date: 2026/09/06
tags:
  - attack.command_and_control
  - attack.t1071.001
  - attack.t1573.001
  - attack.t1001.003
  - attack.t1041
logsource:
  category: network_connection
  product: linux
detection:
  selection_websocket:
    DestinationPort: 443
    DestinationIp|contains:
      - '99.84.67.'
    DestinationHostname|contains:
      - 'windwsecurity.run'
      - '247.cdnflare.xyz'
  selection_ntp:
    DestinationPort: 123
    Protocol: 'UDP'
    DestinationHostname|contains:
      - 'ntp.timesysnc.net'
      - 'time.microsft.run'
      - 'pool.microsft.studio'
  condition: selection_websocket or selection_ntp
falsepositives:
  - Legitimate NTP traffic to correctly-spelled time servers — these domains are typosquats of legitimate services (timesysnc vs timesync, microsft vs microsoft)
  - CDN traffic to Cloudflare IP ranges that include 99.84.x.x — correlate with known Magento storefront CDN configuration
level: high

Known false positives:

  • Legitimate Cloudflare CDN traffic to the same IP range (99.84.x.x) — correlate with known storefront CDN configuration
  • Standard NTP time synchronization to legitimate time servers with similar names

Tuning notes: Monitor the typosquatted NTP domains with particular attention — correctly-spelled equivalents should exist in your organization's allowed NTP server list. The NTP C2 channel is designed to blend with legitimate NTP traffic; inspect packet payload sizes (custom NTP-shaped traffic may have unusual payload lengths vs standard NTP). Block 247.cdnflare.xyz at DNS/network level. Monitor for outbound WebSocket connections to unexpected destinations on port 443, especially from web server hosts.

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)
  • StyleSmuggler Magento Log Poisoning — Detection of PHP Code in Application Logs · unlock with TL Creds
  • StyleSmuggler Backdoor Persistence — Cron Entry and Binary Execution Detection · unlock with TL Creds
  • StyleSmuggler NTP-Shaped C2 Beacon to Typosquatted Domains · unlock with TL Creds
  • StyleSmuggler Two-Stage Exploitation Chain — GraphQL Injection via Template Rendering · unlock with TL Creds
  • StyleSmuggler Rust Backdoor — Persistent Implant Masquerading as Kernel Worker · unlock with TL Creds
  • StyleSmuggler C2 Communication Over NTP-Shaped and DNS Tunnel Channels · unlock with TL Creds
  • StyleSmuggler GraphQL styles[] Parameter Injection — Exploitation Stage · unlock with TL Creds
  • StyleSmuggler Kworker Process Masquerading and Backdoor Execution — Post-Exploitation · unlock with TL Creds

Indicators of compromise

As of 2026-11-07, Threadlinqs Intelligence tracks 82 indicators of compromise for TL-2026-2358.

Sample indicators (3 of 82)

  • IP address: 99.84.67[.]186
  • Domain: 247[.]cdnflare[.]xyz
  • URL: hxxps://www[.]incofar[.]it/js/jquery/plugins/ajaxfileupload/mag.txt

Defanged, TLP:CLEAR. Full set of 82 indicators: unlock with TL Creds.

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

Technical analysis

Actively exploited since September 4, 2026 with confirmed compromises of multiple stores and no official patch or CVE as of publication.

StyleSmuggler is an unauthenticated remote code execution (RCE) zero-day vulnerability affecting Magento Open Source and Adobe Commerce, discovered by Dutch e-commerce security firm Sansec on September 4, 2026. The vulnerability is actively exploited in the wild with confirmed compromises of at least two stores managed by Disrex Group, plus additional victims detected via Sansec's eComscan platform. As of September 6, 2026, no CVE has been assigned by Adobe, no official patch has been released, and all supported versions of Magento Open Source (2.4.6 through 2.4.9) and the corresponding Adobe Commerce versions are confirmed affected — including one victim running 2.4.6-p15 with July and August 2026 security patches applied.

The attack follows a sophisticated two-stage chain. In Stage 1 (injection), the attacker sends a malicious HTTP request through Magento's GraphQL endpoint with crafted styles[] parameters that bypass input sanitization and force Magento's template rendering engine to write PHP code into files the application itself maintains — either var/log/system.log (via an invalid store code logged verbatim) or var/report/ (via uncaught exception failure reports). The trigger header marking the poisoned content is an X-TRACE- followed by ten hex characters (early campaign) or X- followed by twelve hex characters (later same day, after the marker was deliberately changed).

In Stage 2 (execution), the attacker triggers Magento's built-in Payment Transaction Failed Reminder email by submitting an order with a .invalid domain address and a zero total. Magento's getProcessedTemplate() method parses the {{block}} directive embedded in the attacker-controlled text, instantiating objects through parameters like generatorClass and with_resolved. The object-injection chain walks through Magento's internal class hierarchy until it reaches one of three dependency-injection compiler scanner methods — ArrayScanner::collectEntities(), ClassesScanner::includeClass(), or XmlInterceptorScanner::_handleControllerClassName() — in setup/src/Magento/Setup/Module/Di/Code/. These methods accept attacker-controlled file paths and execute them via PHP include or require_once. Because PHP's include executes any PHP content in the file, a log file full of harmless-looking diagnostic messages becomes executable code. A TypeError from array_merge() with an integer argument in system.log immediately after the include is a forensic tell confirming successful exploitation; stealthier variants append return []; to the payload, leaving no error trace.

Upon code execution, a PHP dropper probes six PHP process-execution functions sequentially — shell_exec, exec, system, passthru, proc_open, popen — and uses the first available. On one store where the first four were disabled by open_basedir, proc_open remained enabled and was sufficient for the dropper to spawn the child process, which then operated outside PHP's confinement entirely. The dropper downloads an architecture-matched (~1.9 MB, stripped, statically linked) Rust binary from https://247.cdnflare.xyz/files/kworker-linux-<arch>, compiles for both x86-64 and arm64, makes it executable, and detaches it as a background process.

The binary installs itself at ~/.local/share/.gvfsd/gvfsd-user — outside the web document root, under the site user's home directory — and masquerades as [kworker/u:8:0] to mimic a legitimate Linux kernel thread. Genuine kernel threads are owned by root with zero resident memory; a bracketed process name owned by a site user with real RSS consumption is the definitive detection signal. The implant sets its command-line string to the bracketed value directly, so checks against the comm field (which ps derives from args) match nothing.

Persistence is achieved through a cron entry written directly to /var/spool/cron/crontabs/<user>, bypassing the crontab command so syslog records no REPLACE events. The entry executes the implant every 5 minutes: */5 * * * * exec <home>/.local/share/.gvfsd/gvfsd-user. The implant re-adds the entry within one second of removal; one store exhibited 1,728 identical lines. The binary in memory may differ from the file on disk, as confirmed on one compromised store — defenders are advised to hash the running process from /proc/<pid>/exe as well as the file on disk. The implant persists even after file deletion through the deleted inode accessible via /proc/<pid>/exe.

C2 communication occurs over multiple channels: WebSocket over TLS to 99.84.67.186:443 (and windwsecurity.run:443 for remote shell access), as well as custom NTP-shaped traffic over UDP port 123 to ntp.timesysnc.net, time.microsft.run, and pool.microsft.studio — blending with legitimate NTP traffic on the network. On at least one compromised store, two 200+ MB packet captures contained zero packets to any known C2 address; instead, the implant held 28 simultaneous connections to the local Redis instance on 127.0.0.1:6379, reading Magento session data. This pattern makes egress-based network monitoring unreliable as a compromise indicator.

Attack infrastructure includes 26+ distinct source addresses spanning at least two waves: two hosting-provider IPs (5.181.86.133 from CloudVPS with 96 requests, 91.238.181.19 from AS49434 with 48 requests in a second wave) and 24 residential proxy IPs from consumer ISPs sending 2-6 requests each. Blocking only the most prominent attacker IP (88.216.72.181 with 45 requests) would stop less than a quarter of observed traffic. The malware download host 247.cdnflare.xyz resolves to both IPv4 and IPv6 (2a06:98c1:3120::2, 2a06:98c1:3121::2).

Several early-warning signs help merchants detect compromise: (1) a Payment Transaction Failed Reminder email containing raw unresolved {{var ...}} template tags, a customer address on a .invalid domain, and a total of zero — described by Disrex as exhaust from the exploitation attempt passing through Magento's template filter; (2) Magento's fallback 'an error occurred generating this content' message inside email address blocks; (3) unexpected bursts of Payment Transaction Failed Reminder emails; (4) the presence of X_TRACE_, X_-hex, or <?php markers in var/report/ or var/log/system.log; (5) processes named [kworker] owned by non-root users with non-zero resident memory; (6) cron entries referencing gvfsd or .kw_ in /var/spool/cron/crontabs/; and (7) Response payloads wrapped in MG<20hex>::<base64>::/MG<20hex> patterns.

Multiple unofficial mitigations have been published. Disrex Group shipped a composer-patches source patch adding a PHP_SAPI !== 'cli' guard to all three DI scanner methods, verified working on Magento 2.4.6 through 2.4.9 and live-tested on 2.4.7-p2 and 2.4.8-p4. ProxiBlue (Lucas van Staden) independently published the identical guard on September 5. Graycore published a Composer-installable hardening module (graycore/magento2-style-smuggler-patch) with three entry-point protections: email template block directive filtering, grid row URL generator class validation, and PHP open tag breaking in Web API fatal error reports — explicitly noting this is hardening, not a fix. Server-level protections include disabling all six PHP process-execution functions (especially proc_open), mounting /tmp, /var/tmp, and /dev/shm with noexec, and deploying web server rules blocking styles[], generatorClass, and with_resolved parameters in query strings (bypassable via POST body). Temporarily disabling GraphQL is recommended for classic/Hyvä storefronts that do not require it. Sansec Shield detection rules have been operational since September 5 at 07:15 UTC. Adobe's next scheduled security release is September 8, 2026, although it has not been confirmed to address this vulnerability.

MITRE ATT&CK techniques by tactic:

Command and Control

T1001.003 Protocol or Service Impersonation; T1008 Fallback Channels; T1071.001 Web Protocols; T1071.004 Application Layer Protocol: DNS; T1090 Proxy; T1090.002 Proxy: External Proxy; T1095 Non-Application Layer Protocol; T1105 Ingress Tool Transfer; T1573 Encrypted Channel; T1573.001 Encrypted Channel: Symmetric Cryptography

Collection

T1005 Data from Local System

Defense Evasion

T1027 Obfuscated Files or Information; T1036.004 Masquerading: Masquerade Task or Service; T1036.005 Match Legitimate Resource Name or Location; T1070.004 Indicator Removal: File Deletion; T1140 Deobfuscate/Decode Files or Information; T1497.001 Virtualization/Sandbox Evasion: System Checks; T1564.001 Hide Artifacts: Hidden Files and Directories; T1622 Debugger Evasion

Exfiltration

T1041 Exfiltration Over C2 Channel; T1048.003 Exfiltration Over Unencrypted Non-C2 Protocol; T1567 Exfiltration Over Web Service

Persistence

T1053.003 Scheduled Task/Job: Cron; T1505.003 Server Software Component: Web Shell; T1543.002 Create or Modify System Process: Systemd Service

Execution

T1059 Command and Scripting Interpreter; T1059.004 Command and Scripting Interpreter: Unix Shell; T1059.006 Command and Scripting Interpreter: Python; T1059.007 JavaScript

Discovery

T1082 System Information Discovery; T1083 File and Directory Discovery

Initial Access

T1190 Exploit Public-Facing Application

Resource Development

T1583.001 Acquire Infrastructure: Domains; T1587.001 Develop Capabilities: Malware; T1608.001 Stage Capabilities: Upload Malware

Reconnaissance

T1595.002 Active Scanning: Vulnerability Scanning

Remediation

Patches

  • Disrex source patch: CLI-only guard on ArrayScanner::collectEntities(), ClassesScanner::includeClass(), XmlInterceptorScanner::_handleControllerClassName() — verified 2.4.6 through 2.4.9
  • Graycore magento2-style-smuggler-patch: Composer module hardening email template block directives, grid URL generation, and Web API error report injection

Immediate actions

  • Disable GraphQL endpoint if not required by storefront (classic and Hyvä themes; exclude headless/PWA)
  • Add all six PHP process-execution functions (shell_exec, exec, system, passthru, proc_open, popen) to disable_functions
  • Mount /tmp, /var/tmp, and /dev/shm with noexec to block binary execution from writable directories
  • Deploy web server rules blocking styles[], generatorClass, with_resolved, and {{ in query strings (note: POST body bypasses these)
  • Deploy Sansec Shield or equivalent WAF rules covering both stages of the exploit chain
  • Run eComscan 1.9.7+ with widened scan path (include site user home directory, not just document root)

Workarounds

  • Temporarily disable GraphQL for classic and Hyvä storefronts
  • Add nginx/Apache query-string filters blocking styles[, generatorClass, with_resolved, {{, <? — POST body bypasses these
  • Block 247.cdnflare.xyz and known C2 domains at DNS/network level
  • Block known attacker IPs at firewall (88.216.72.181, 5.181.86.133, 91.238.181.19) — residential proxy pool cannot be fully blocked

Longer-term hardening

  • Apply Disrex composer-patches source patch or ProxiBlue patch to guard DI scanner methods with PHP_SAPI !== 'cli'
  • Deploy Graycore hardening module (graycore/magento2-style-smuggler-patch) for template/grid/error protection
  • Implement process monitoring for [kworker] process names owned by non-root users with non-zero RSS
  • Monitor for Payment Transaction Failed Reminder email bursts and malformed {{var ...}} template tags
  • Set up cron file integrity monitoring on /var/spool/cron/crontabs/
  • Schedule regular Redis session store audits for unexpected connection counts
  • Apply official Adobe patch when released and verify composership integration

Affected and fixed versions

  • Adobe — Adobe Commerce
    Vulnerable versions: 2.4.6; 2.4.7; 2.4.8; 2.4.9; prior releases
  • Magento — Magento Open Source
    Vulnerable versions: 2.4.6; 2.4.7; 2.4.8; 2.4.9; prior releases

Timeline

Last updated · 7 updates · revalidated 7× · latest source

Showing the 20 most recent tracked events.

  • Store B (Magento 2.4.7-p2) compromised at 00:55 UTC; both stores breached within the ~8-hour window before any defense existed
  • Implant variant v2.1.4 released: binary renamed to fc-cache, stored at locked-ioc-25; C2 traffic shifted from TLS/WebSocket to NTP-disguised UDP on port 123
  • 17:30 UTC — Sansec blocks a GraphQL recon probe (storeConfig { store_code }) against a 2.4.7-p10 storefront; the probe carried php_uname()/get_current_user()/getcwd() and a pub/media-writability check in the Store: header, exfiltrated as 50-char DNS-subdomain labels.
  • BleepingComputer, SecurityWeek, heise online, and Aikido publish articles on StyleSmuggler; Adobe Enterprise Support confirms working on a fix
  • Adobe publishes emergency hotfix APSB26-146 (VULN-39341, CVE-2026-75650) via repo.magento.com at 20:20 UTC; tested on 2.4.4-2.4.9 August releases and Commerce B2B 1.3.3-1.5.3
  • Implant variant v2.1.5 released: binary renamed to chronyd, stored at locked-ioc-26
  • A second, unrelated attacker deploys a 485-byte PHP web shell to pub/media/ via StyleSmuggler and exfiltrates data via DNS to an OAST service on oastify.com
  • Scandiweb backports a community security patch covering CVE-2026-75650 to 41 legacy/unsupported Magento releases (2.2, 2.3, 2.4.0-2.4.3).
  • A public proof-of-concept exploit repository for CVE-2026-75650 appears, per CrowdSec tracking, accelerating opportunistic scanning.
  • CISA adds CVE-2026-75650 (StyleSmuggler) to the Known Exploited Vulnerabilities Catalog; per BOD 26-04, federal agencies must remediate by September 11, 2026.
  • NVD publishes the CVE-2026-75650 record with CVSS 10.0; CISA KEV had not yet listed the vulnerability as of publication.
  • Adobe's next scheduled security release — not confirmed to address StyleSmuggler
  • Start of the Imperva-tracked coordinated campaign delivering locked-ioc-58, the chi_down Go RAT and the locked-ioc-59 webshell against Magento stores in Western Europe, North America and Asia-Pacific.
  • CrowdSec deploys a crowdsourced detection rule for CVE-2026-75650 exploitation traffic; daily exploitation signals average 552/day through September 13.
  • Cloudflare and Imperva deploy WAF rules covering StyleSmuggler exploitation.
  • Exploitation attempts peak at 1,303 signals in a single day from 193 distinct sources, per CrowdSec telemetry.
  • End of the Imperva-observed activity window: five attack clusters and 96 locked-ioc-58 downloads across 12 payload variants, with 0% VirusTotal detection at analysis time.
  • Akamai publishes a detailed public technical write-up of CVE-2026-75650 ("StyleSmuggler"), naming it and detailing the PayPal-endpoint injection vector, WraithC2, and gs-netcat tooling.
  • Imperva publishes its analysis of the PSK Agent RAT campaign, including an assessment of a likely Chinese-speaking developer/operator at moderate confidence.
  • Cyber Security News reports the PSK Agent RAT deployment and the Imperva findings.

Update history

Sources cited

Other reports on CVE-2026-75650

Previous and next in vulnerability

Drafted by the Threadlinqs research pipeline from 11 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