ARTEX Automated Endpoint and Network Maneuvering
An automated attacker is coordinating dual-use offensive tools like VShell, Chisel, or fscan to perform high-velocity internal discovery and credential harvesting.
Based on research by Cisco Talos 2026-11-14 9 steps · 3 queries T1003.001 T1046 T1105 T1555 T1572
Brief
Why Now
Adversaries move from manual command-line interaction toward agentic frameworks. The Cisco Talos report, ARTEX and beyond: AI-enabled penetration testing platforms, describes how these systems use Large Language Models and automated planners to coordinate worker agents. These agents execute tasks like internal reconnaissance and credential theft much faster than a human operator. This shift requires a move from detecting individual tools to identifying the machine-speed orchestration of those tools. Traditional rules might catch a single binary, but they often miss the coordinated velocity that defines these AI-enabled platforms.
How the Hunt Flows
The first phase identifies the execution of binaries associated with ARTEX worker agents. The query scans process activity for specific filenames and original filenames like fscan.exe, vshell.exe, nps.exe, and adaptix.exe. This step scopes the investigation to hosts where offensive toolsets are present, providing the initial leads for behavioral correlation. We focus on known filenames identified in the worker agent workflows.
Once the hunt identifies a lead host, it pivots into two parallel investigations. The first examines network telemetry to find processes contacting a high volume of unique internal destination IPs in a short window. This identifies automated internal scanning. The second investigation looks for module activity where non-system processes load lsasrv.dll or samlib.dll, indicating an attempt to harvest credentials from memory. These two streams run simultaneously to gather evidence of the agent's next steps in the attack chain.
The final phase assesses whether these events indicate automated coordination. An analyst or an automated triage agent reviews the findings to see if a single host exhibited both high-velocity scanning and credential access attempts shortly after an offensive tool execution. This correlation distinguishes an automated framework from a human operator who typically moves with lower velocity and higher latency between actions. The presence of multiple indicators in a tight temporal window serves as the primary signal for orchestration.
What the Hunt Cannot See
This hunt faces two primary blind spots. First, if an automated agent performs reconnaissance on internal network segments where the endpoint sensor lacks visibility, the prevalence baseline will not trigger. This risk is higher in environments with restricted internal monitoring. Second, tools executed entirely in memory via shellcode or reflective loading may bypass the initial file-based process lead. In these cases, the hunt relies solely on the behavioral module and network queries, which may be more prone to noise if the process name is common or missing.
In this series
Steps
-
Offensive Tool Execution Lead
Query · scopingIdentify hosts executing binaries associated with ARTEX worker agents to narrow the search space.
reads hb_process_activitysqlSELECT device_hostname, process_name, process_cmd_line, process_original_file_name, user_name, time FROM hb_process_activity WHERE (instr(',' || '{{offensive_tools}}' || ',', ',' || LOWER(process_name) || ',') > 0 OR instr(',' || '{{offensive_tools}}' || ',', ',' || LOWER(process_original_file_name) || ',') > 0) AND time >= datetime('now', '-{{lookback_days}} days')What a hit looks like. Rows indicating the execution of known offensive tools. Silence suggests no direct matches for the named binaries occurred.
-
Network Discovery Prevalence
Query · baselineDetect automated internal scanning by finding processes that contact a high volume of unique internal destination IPs.
reads hb_network_connectionsqlSELECT device_hostname, process_name, COUNT(DISTINCT dst_endpoint_ip) AS unique_targets, COUNT(*) AS total_conns FROM hb_network_connection WHERE ('{{scope_hosts}}' = '' OR instr(',' || '{{scope_hosts}}' || ',', ',' || device_hostname || ',') > 0) AND time >= datetime('now', '-{{lookback_days}} days') GROUP BY device_hostname, process_name HAVING unique_targets > 15 ORDER BY unique_targets DESCWhat a hit looks like. Processes (like fscan or vshell) that target dozens of internal hosts in a short window. Genuine admin tools may show this, but rare process names should stand out.
-
Credential Harvesting Triage
Query · triageCorroborate tool execution by finding module loads related to LSASS or SAM access from non-system directories.
reads hb_module_activitysqlSELECT device_hostname, process_name, module_path, module_signature_subject, time FROM hb_module_activity WHERE (LOWER(module_path) LIKE '%\lsasrv.dll' OR LOWER(module_path) LIKE '%\samlib.dll') AND NOT (LOWER(process_name) LIKE 'c:\windows\system32\%' OR LOWER(process_name) LIKE 'c:\windows\syswow64\%') AND ('{{scope_hosts}}' = '' OR instr(',' || '{{scope_hosts}}' || ',', ',' || device_hostname || ',') > 0) AND time >= datetime('now', '-{{lookback_days}} days')What a hit looks like. Any row where a user-space binary is interacting with the credential databases. Silence proves the tools did not attempt LSASS memory access within the window.
-
Triage Orchestration
Agent triageAssess whether the observed tools and behaviors indicate an automated framework coordination.
-
Route on Verdict
DecisionDirect the hunt to containment or manual review.
-
Isolate Affected Endpoint
Response actionDisrupt the automated agent's planner by isolating the host.
-
Analyst Validation
Analyst taskConfirm the findings and assess for the presence of the agent harness.
-
Hunt Close-out
Analyst taskDocument findings and ensure EDR health across the fleet.
Coverage
Scenario coverage
| Stage | Covered | How, or why not |
|---|---|---|
| Orchestrated C2 and Tunneling T1572 |
Yes | offensive-tool-lead, network-discovery-prevalence |
| Internal Discovery Automation T1046 |
Yes | network-discovery-prevalence |
| Automated Credential Harvesting T1003.001 · T1555 |
Yes | credential-harvesting-triage |
| Automated Vulnerability Probing T1190 |
Out of scope | Belongs to another part of the 'ARTEX and beyond: AI-enabled penetration testing platforms' series. |
| High-Velocity Credential Spraying T1110.003 |
Out of scope | Belongs to another part of the 'ARTEX and beyond: AI-enabled penetration testing platforms' series. |
| Cloud Privilege Manipulation T1098 |
Out of scope | Belongs to another part of the 'ARTEX and beyond: AI-enabled penetration testing platforms' series. |
Blind spots
- Needs hb_network_connection for all internal segments. An automated agent could perform reconnaissance on unmonitored subnets without triggering the prevalence baseline. It would answer whether fscan is scanning segments not captured by endpoint sensors.
- Needs hb_process_activity (on_disk = 0) and hb_module_activity. Tools executed entirely in memory bypass the file-name lead query and are only visible if the module load or process tampering is successfully audited. It would answer whether tools like Mimikatz were loaded directly into memory via shellcode.
Parameters & data
Parameters
| Parameter | Type | Default | What it is |
|---|---|---|---|
lookback_days | number | 14 | Days of historical telemetry to examine. |
offensive_tools | list[string] | fscan.exe, vshell.exe, chisel.exe, nps.exe, mimikatz.exe, adaptix.exe, gophish.exe | Names of offensive or dual-use binaries identified in ARTEX worker agent workflows. |
scope_hosts | list[host] | — | List of hostnames from the scoping step to narrow the behavioral queries; leave empty to hunt across the whole estate. |
Telemetry
| Source | Category | Telemetry |
|---|---|---|
| Endpoint telemetry (hb_ surfaces) | endpoint | endpoint |
| Network telemetry | network | network |
Source
---
analysis: While a rule can fire on a specific binary name, this hunt provides the
context of machine-speed network scanning and LSASS module interaction, allowing
an analyst to identify automated coordination rather than a manual pentest.
blind_spots:
- id: limited-network-visibility
question: whether fscan is scanning segments not captured by endpoint sensors
requires: hb_network_connection for all internal segments
risk: An automated agent could perform reconnaissance on unmonitored subnets without
triggering the prevalence baseline.
stage: internal-discovery-automation
- id: in-memory-tool-execution
question: whether tools like Mimikatz were loaded directly into memory via shellcode
requires: hb_process_activity (on_disk = 0) and hb_module_activity
risk: Tools executed entirely in memory bypass the file-name lead query and are
only visible if the module load or process tampering is successfully audited.
stage: automated-credential-harvesting
coverage:
- stage: orchestrated-c2-and-tunneling
status: covered
steps:
- offensive-tool-lead
- network-discovery-prevalence
- stage: internal-discovery-automation
status: covered
steps:
- network-discovery-prevalence
- stage: automated-credential-harvesting
status: covered
steps:
- credential-harvesting-triage
- reason: 'Belongs to another part of the ''ARTEX and beyond: AI-enabled penetration
testing platforms'' series.'
stage: automated-vulnerability-probing
status: out_of_scope
- reason: 'Belongs to another part of the ''ARTEX and beyond: AI-enabled penetration
testing platforms'' series.'
stage: high-velocity-credential-spraying
status: out_of_scope
- reason: 'Belongs to another part of the ''ARTEX and beyond: AI-enabled penetration
testing platforms'' series.'
stage: cloud-privilege-manipulation
status: out_of_scope
guardrails:
claims: no_unsupported
evidence: citation_required
missing_data: not_benign
telemetry: untrusted
hunt:
applicability: campaign-specific
handoff: keep-as-periodic-hunt
justification: AI-enabled frameworks like ARTEX can move from entry to credential
theft in minutes; a negative result over the estate confirms no machine-speed
maneuvering occurred within the retention window.
methodology: model-assisted
trigger: intel-report
hypothesis: An automated attacker is coordinating dual-use offensive tools like VShell,
Chisel, or fscan to perform high-velocity internal discovery and credential harvesting.
labels:
- hunt
- attack.t1572
- attack.t1046
- attack.t1003.001
- attack.t1555
- attack.t1105
- command and control
- credential access
- discovery
- initial access
- persistence
name: ARTEX Automated Endpoint and Network Maneuvering
parameters:
lookback_days:
default: '14'
description: Days of historical telemetry to examine.
type: number
offensive_tools:
default:
- fscan.exe
- vshell.exe
- chisel.exe
- nps.exe
- mimikatz.exe
- adaptix.exe
- gophish.exe
description: Names of offensive or dual-use binaries identified in ARTEX worker
agent workflows.
from:
kind: article
observed: '2026-11-13'
ref: https://blog.talosintelligence.com/artex-and-beyond/
type: list[string]
scope_hosts:
default: []
description: List of hostnames from the scoping step to narrow the behavioral
queries; leave empty to hunt across the whole estate.
type: list[host]
provenance:
authors:
- name: Huntbase hunt generation
org: huntbase.io
generated:
by: huntbase-hunt-generation
from: https://blog.talosintelligence.com/artex-and-beyond/
gates:
- dry-run
- lint
- critic
model: hb_google/gemini-3-flash-preview
rationale: Start with servers and high-value workstation segments. If the lead query
is empty, consider widening the offensive_tools list to include generic names like
'scan.exe' or 'tunnel.exe'.
references:
- name: "Cisco Talos \u2014 ARTEX and beyond: AI-enabled penetration testing platforms"
url: https://blog.talosintelligence.com/artex-and-beyond/
related:
- hunt: automated-vulnerability-probing
reason: The initial stage of vulnerability probing and web shell deployment is a
separate hunt focused on web server logs and public-facing exposures.
relation: out-of-scope-alternative
- hunt: ai-driven-identity-cloud-exploitation
relation: follows
scenario:
stages:
- name: Automated Vulnerability Probing
observables:
- Acunetix web scanner
- Nessus vulnerability scanner
- Interactsh out-of-band detection
- Inbound scanning traffic from ARTEX-associated C2 IPs
slug: automated-vulnerability-probing
tactic: initial-access
techniques:
- T1190
- name: High-Velocity Credential Spraying
observables:
- Accelerated sign-in attempts against downstream Microsoft 365 tenants
- Compromised ConnectWise application credentials used for multi-tenant access
slug: high-velocity-credential-spraying
tactic: credential-access
techniques:
- T1110.003
- name: Orchestrated C2 and Tunneling
observables:
- VShell C2 framework
- AdaptixC2 framework
- Chisel TCP/UDP tunneling
- Net Proxy Server (NPS) tunneling
- multios.trojan.adaptixc2 signature
slug: orchestrated-c2-and-tunneling
tactic: command-and-control
techniques:
- T1572
- name: Internal Discovery Automation
observables:
- fscan internal network scanner
- multios.tool.fscan signature
slug: internal-discovery-automation
tactic: discovery
techniques:
- T1046
- name: Automated Credential Harvesting
observables:
- Mimikatz execution
- LSASS memory access
- Credential leak management services integration
slug: automated-credential-harvesting
tactic: credential-access
techniques:
- T1003.001
- T1555
- name: Cloud Privilege Manipulation
observables:
- Microsoft Graph API activity
- Entra ID application credential changes
- Role assignment and privilege modifications
slug: cloud-privilege-manipulation
tactic: persistence
techniques:
- T1098
summary: Threat actors are leveraging AI-driven penetration testing frameworks like
ARTEX and PentAGI to automate reconnaissance, credential spraying, and post-exploitation
at machine speed. These frameworks orchestrate a variety of offensive tools including
C2 frameworks (VShell, AdaptixC2), tunneling utilities (Chisel, NPS), and internal
scanners (fscan) to rapidly compromise targets across both endpoint and cloud
environments.
series:
index: 2
slug: artex-and-beyond-ai-enabled-penetration-testing-platforms
title: 'ARTEX and beyond: AI-enabled penetration testing platforms'
total: 2
severity: high
targets:
analyst:
name: Tier-2 analyst
role: analyst
endpoint:
category: endpoint
name: Endpoint telemetry (hb_ surfaces)
telemetry:
- endpoint
hunter:
agent: true
name: Hunt agent
network:
category: network
name: Network telemetry
telemetry:
- network
tlp: clear
type: investigation
---
# ARTEX Automated Endpoint and Network Maneuvering
ARTEX and similar agentic frameworks coordinate worker agents to deploy offensive toolsets at machine speed. This hunt identifies the execution of common binaries used by these agents—including VShell, Chisel, and fscan—and correlates them with behavioral indicators of automated network scanning and unauthorized module loads into LSASS. By examining these signals across process, network, and module telemetry, an analyst can distinguish between a human operator and an objective-driven automated framework.
## offensive-tool-lead
<!-- Offensive Tool Execution Lead -->
Identify hosts executing binaries associated with ARTEX worker agents to narrow the search space.
```sqlite target=endpoint role=scoping params=(offensive_tools=offensive_tools, lookback_days=lookback_days)
~~~yaml
expected: Rows indicating the execution of known offensive tools. Silence suggests
no direct matches for the named binaries occurred.
reads:
- device_hostname
- process_name
- process_cmd_line
- process_original_file_name
- user_name
- time
silence: not_evidence_of_absence
source: hb_process_activity
verified: dry-run
verified_at: '2026-11-14'
~~~
SELECT device_hostname, process_name, process_cmd_line, process_original_file_name, user_name, time FROM hb_process_activity WHERE (instr(',' || '{{offensive_tools}}' || ',', ',' || LOWER(process_name) || ',') > 0 OR instr(',' || '{{offensive_tools}}' || ',', ',' || LOWER(process_original_file_name) || ',') > 0) AND time >= datetime('now', '-{{lookback_days}} days')
```
## maneuver-fan-out
<!-- Maneuver Fan-out -->
parallel:
- → network-discovery-prevalence
- → credential-harvesting-triage
join: → triage-orchestration
## network-discovery-prevalence
<!-- Network Discovery Prevalence -->
Detect automated internal scanning by finding processes that contact a high volume of unique internal destination IPs.
```sqlite target=network role=baseline params=(lookback_days=lookback_days, scope_hosts=scope_hosts)
~~~yaml
baseline:
compare: new_this_window
window: '{{lookback_days}}d'
expected: Processes (like fscan or vshell) that target dozens of internal hosts in
a short window. Genuine admin tools may show this, but rare process names should
stand out.
prevalence:
by: device_hostname
key:
- process_name
rare_below: 3
reads:
- device_hostname
- process_name
- dst_endpoint_ip
- time
silence: not_evidence_of_absence
source: hb_network_connection
verified: dry-run
verified_at: '2026-11-14'
~~~
SELECT device_hostname, process_name, COUNT(DISTINCT dst_endpoint_ip) AS unique_targets, COUNT(*) AS total_conns FROM hb_network_connection WHERE ('{{scope_hosts}}' = '' OR instr(',' || '{{scope_hosts}}' || ',', ',' || device_hostname || ',') > 0) AND time >= datetime('now', '-{{lookback_days}} days') GROUP BY device_hostname, process_name HAVING unique_targets > 15 ORDER BY unique_targets DESC
```
## credential-harvesting-triage
<!-- Credential Harvesting Triage -->
Corroborate tool execution by finding module loads related to LSASS or SAM access from non-system directories.
```sqlite target=endpoint role=triage params=(lookback_days=lookback_days, scope_hosts=scope_hosts)
~~~yaml
expected: Any row where a user-space binary is interacting with the credential databases.
Silence proves the tools did not attempt LSASS memory access within the window.
reads:
- device_hostname
- process_name
- module_path
- module_signature_subject
- time
silence: not_evidence_of_absence
source: hb_module_activity
verified: dry-run
verified_at: '2026-11-14'
~~~
SELECT device_hostname, process_name, module_path, module_signature_subject, time FROM hb_module_activity WHERE (LOWER(module_path) LIKE '%\lsasrv.dll' OR LOWER(module_path) LIKE '%\samlib.dll') AND NOT (LOWER(process_name) LIKE 'c:\windows\system32\%' OR LOWER(process_name) LIKE 'c:\windows\syswow64\%') AND ('{{scope_hosts}}' = '' OR instr(',' || '{{scope_hosts}}' || ',', ',' || device_hostname || ',') > 0) AND time >= datetime('now', '-{{lookback_days}} days')
```
## triage-orchestration
<!-- Triage Orchestration -->
```agent target=hunter
cite: required
context:
- offensive-tool-lead
- network-discovery-prevalence
- credential-harvesting-triage
max_iterations: 5
objective: Determine if the execution of offensive tools, the high-velocity network
connections, and the LSASS module loads together indicate an automated AI-enabled
framework like ARTEX on any single host.
success_criteria: A verdict of malicious, suspicious, or benign per host, citing the
offensive binary name and connection volume.
tools:
- endpoint
- network
```
## route-on-verdict
<!-- Route on Verdict -->
if~: "the triage verdict is malicious for at least one host involving an offensive tool binary" (confidence: high, judge=hunter)
then: → isolate-endpoint
indeterminate: → analyst-validation
unavailable: → analyst-validation (blind_spot: limited-network-visibility)
else: → hunt-closeout
## isolate-endpoint
<!-- Isolate Affected Endpoint -->
```action target=endpoint
~~~yaml
approval: required
~~~
Isolate the host and terminate any processes identified by the triage agent as malicious.
```
→ analyst-validation
## analyst-validation
<!-- Analyst Validation -->
```manual target=analyst
Review the process ancestry for the offensive binaries; search for unusual Python or Node.js processes that may be acting as the ARTEX agent harness coordinating the tools.
```
→ hunt-closeout
## hunt-closeout
<!-- Hunt Close-out -->
```manual target=analyst
Record the outcome and verify that all hosts in the affected segment have reporting EDR agents, as ARTEX frameworks often include bypass capabilities.
```
→ end
Run it
Take this hunt into your environment.
Open it in Huntbase to run every step against your own connections, with Scout weighing the evidence and your analysts in command. Or take the open hunt.md file anywhere that reads the format.
Machine-drafted by huntbase-hunt-generation using hb_google/gemini-3-flash-preview, gated by dry-run, lint, critic, then reviewed by a person.