# ARTEX open-source agentic AI pentest framework abused by Chinese-nexus state and criminal actors to automate intrusions at machine speed (FBI Flash Alert)

> An FBI flash alert warns that ARTEX, an open-source agentic AI penetration-testing framework that can drive any LLM, is being used by cybercriminals and nation-state actors to automate intrusions from reconnaissance through data exfiltration, with some intrusions reaching privileged access in under three hours. Microsoft ties it to Storm-0940's ConnectWise Cloud Backup supply-chain compromise; CrowdStrike, Amazon, Check Point and GMO document further campaigns.

- **Published:** 2026-11-13T22:04:57Z
- **Last reviewed:** 2026-11-14T07:25:40.849Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-6001
- **ID:** TL-2026-6001
- **Severity:** HIGH
- **Category:** THREAT_INTEL
- **Status:** ACTIVE
- **Actor:** Storm-0940 (China)
- **Detections:** 9 (1 shown free below) · **IOCs:** 48 (3 shown free below)
- **CVEs:** CVE-2025-3248, CVE-2026-28797

## Description

ARTEX (also written Artex) is an open-source, multi-agent penetration-testing framework written in Go with a Next.js web console and PostgreSQL backend. It was published to GitHub as Autumn-27/ARTEX under AGPL-3.0 on 2026-07-26 (Censys; some press reports say August). A Planner agent turns a high-level goal into attack 'intents', Worker agents claim intents, run tools (the Docker image ships with nmap) and write results back, and verification/compression agents plus a persistent asset/exploration knowledge graph let multiple sessions run in parallel with minimal human input. It integrates Anthropic, OpenAI and OpenAI-compatible endpoints and has been used with DeepSeek, GLM, Grok and Kimi models; the FBI found Chinese models, especially DeepSeek Flash-family models, most prevalent in attacker-controlled instances because of capability, low cost and no observable guardrails when framed as pentesting. After abuse reports the developer posted a statement on 2026-10-08 disclaiming responsibility and moved the project to closed source; the repository was later removed, but the FBI says the framework persists via community forks, localized variants, archived copies and a commercial version. Sophos additionally reports a trojanized ARTEX Docker image (Troj/Bckdr-SDA), so unofficial mirrors should be treated as untrusted.

The FBI flash alert (as reported by The Record on 2026-11-13; the alert text itself was not retrieved) says ARTEX automates multiple intrusion phases at machine speed, lets low-skill actors run multi-target operations, and has been used against government agencies and dozens of critical sectors worldwide. Some intrusions reached privileged access in under three hours; Microsoft independently measured under three hours from initial targeting to privileged cloud access in several ARTEX-assisted intrusions observed since September. The FBI notes ARTEX generates high volumes of attempts that can trigger defensive sensors, and that actors map trust relationships and compromise weak third-party vendors and less-secured systems (partner/supplier portals, mobile backends, employee apps, sales-support systems, legacy services, third-party integrations).

Reported campaigns: (1) Microsoft attributes ARTEX use to Storm-0940, a China state-aligned group active since 2021, reaching more than 2,250 Microsoft 365 tenants (up to 2,900 sign-in attempts per minute) via cross-tenant password spraying and abuse of OAuth application identities, including the ConnectWise Cloud Backup compromise around 2026-10-14/15 that led ConnectWise to suspend Cloud Backup and related services; Microsoft advises revoking tokens, refresh tokens, sessions and OAuth grants tied to Cloud Backup. (2) CrowdStrike (2026-10-07) documents a likely financially motivated, Chinese-speaking unnamed actor breaching a broker-facing loan inquiry service at one South Korean bank and an employee mobile work-support system at another, using ARTEX configured with a Chinese-language pentest prompt in CLAUDE.md, DeepSeek as primary backend with GLM and Grok supplementary, an LLM API reseller (xcai.pro) and nine proxy IPs. (3) GMO Cybersecurity by Ierae (2026-10-19) found one attack infrastructure targeting about 20 Japanese companies (insurance, finance, logistics, e-commerce, retail, education), with Android/Flutter app reverse engineering (jadx, Blutter) and SSRF testing; operators are unknown and ARTEX's Chinese origin does not by itself establish Chinese or state-backed operators. (4) Amazon Threat Intelligence (MadPot telemetry, 2026-10-03 to 2026-11-05, moderate confidence in ARTEX use and Chinese-nexus attribution) observed AI-model-credential theft that funded further key hunting via LLM gateways and AI app builders (One API, New API, Veloera, Sub2API, LiteLLM, Dify, FastGPT, Langflow, RAGFlow, Open WebUI, LobeChat, Ollama; CVE-2025-3248, CVE-2026-28797, default credentials, SSRF to cloud metadata, secrets in client-side JS) and a separate deepfake-based KYC-bypass effort against two cryptocurrency exchanges. (5) Check Point Research (2026-11-05) found criminal operators using ARTEX against iOS app-signing services (300+ certificates obtained), 13 gaming-account marketplaces (190,000+ Honor of Kings credentials) and 292 crypto-related domains (exposed AWS credentials used to alter S3-hosted iOS distribution assets).

Exposure: Censys counted 1,465 hosts with an exposed ARTEX console around 2026-11-03/04 (about 815 distinct-looking deployments after removing two identical hosting blocks; US 726, Hong Kong 387, China 177, Japan 51, Singapore 38, South Korea 21), default port 8787/tcp; Check Point reports 2,557 systems that hosted panels, 1,608 live. The FBI/The Record cite over 1,600 exposed console instances. Counts differ by scan date and deduplication, and some hosts may be authorized testers. Censys also saw Cobalt Strike, Adaptix C2, Vshell, FRP, NPS, CyberStrikeAI, PentAGI and HexStrike AI on ARTEX hosts. Defender note: ARTEX itself is dual-use; presence of the tool does not prove malicious intent, and attribution to a specific operator or to the Chinese state applies only where a vendor explicitly states it. No hashes were published in the sources reviewed; the BeaconBeagle lookups for 38.244.50.120 and 144.172.64.60 failed (HTTP 502) and should be re-run by the correlation phase.

## MITRE ATT&CK

- T1595.002 Vulnerability Scanning
- T1588.007 Artificial Intelligence
- T1583.003 Virtual Private Server
- T1190 Exploit Public-Facing Application
- T1199 Trusted Relationship
- T1078.004 Cloud Accounts
- T1098.001 Additional Cloud Credentials
- T1505.003 Web Shell
- T1110.003 Password Spraying
- T1552.001 Credentials In Files
- T1550.001 Application Access Token
- T1090 Proxy
- T1657 Financial Theft
- T1593.003 Search Open Websites/Domains
- T1596.003 Search Open Technical Databases
- T1136.003 Create Account
- T1552.005 Unsecured Credentials
- T1528 Steal Application Access Token
- T1580 Cloud Infrastructure Discovery
- T1526 Cloud Service Discovery
- T1619 Cloud Storage Object Discovery
- T1530 Data from Cloud Storage
- T1656 Impersonation
- T1496.004 Resource Hijacking

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**ARTEX-Style Cross-Tenant Entra ID Password Spray (Machine-Speed Failed Sign-ins)** (`TL-2026-6001-SPL-001`)

Detects password-spray behaviour against Microsoft Entra ID / Microsoft 365 characteristic of ARTEX-assisted Storm-0940 activity (TL-2026-6001): one source IP generating failed sign-ins against 10+ distinct accounts within 5 minutes, at machine speed (Microsoft observed up to 2,900 attempts/minute), frequently from scripted user agents. Maps to T1110.003 (Password Spraying) and T1078.004 (Cloud Accounts). ARTEX is dual-use; this rule keys on behaviour, not on the tool.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1110.003, T1078.004
- **Data sources:** azure:aad:signin (Splunk Add-on for Microsoft Cloud Services), SigninLogs, AADNonInteractiveUserSignInLogs, Microsoft Entra ID sign-in logs
- **Severity:** high
- **Confidence:** high

### Splunk SPL

```spl
index=azure sourcetype="azure:aad:signin" earliest=-60m ("status.errorCode"=50126 OR "status.errorCode"=50053 OR "status.errorCode"=50034 OR "status.errorCode"=50057) | bin _time span=5m | stats count AS failures dc(userPrincipalName) AS distinct_users values(userAgent) AS user_agents values(appDisplayName) AS apps by _time ipAddress | where distinct_users>=10 AND failures>=20 | eval scripted_ua=if(match(mvjoin(user_agents,"|"),"(?i)(python-requests|python-httpx|aiohttp|go-http-client|curl/|axios|node-fetch|okhttp)"),"yes","no") | eval attempts_per_min=round(failures/5,1) | sort - failures | table _time ipAddress failures distinct_users attempts_per_min scripted_ua user_agents apps
```

### Microsoft KQL

```kql
union SigninLogs, AADNonInteractiveUserSignInLogs | where TimeGenerated > ago(1h) | where ResultType in ("50126","50053","50034","50057") | summarize Failures=count(), DistinctUsers=dcount(UserPrincipalName), UserAgents=make_set(UserAgent, 5), Apps=make_set(AppDisplayName, 5) by IPAddress, bin(TimeGenerated, 5m) | where DistinctUsers >= 10 and Failures >= 20 | extend ScriptedUA = tostring(UserAgents) matches regex @"(?i)(python-requests|python-httpx|aiohttp|go-http-client|curl/|axios|node-fetch|okhttp)" | extend AttemptsPerMin = round(todouble(Failures) / 5, 1) | project TimeGenerated, IPAddress, Failures, DistinctUsers, AttemptsPerMin, ScriptedUA, UserAgents, Apps
```

### Sigma

```yaml
title: Entra ID Failed Sign-in Base - Password Spray Indicator
name: tl_6001_entra_failed_signin
id: 6001a000-0001-4000-8000-000000000001
status: experimental
description: Failed Entra ID sign-ins with invalid-password, locked, disabled or bad-credential result codes. Consumed by the correlation rule below to detect ARTEX-style machine-speed password spraying (T1110.003).
references:
    - https://therecord.media/artex-ai-tool-fbi-flash-alert
    - https://www.silicon.co.uk/cybersecurity/microsoft-connectwise-artex-631988
author: Threadlinqs Team
date: 2026-11-13
tags:
    - attack.credential-access
    - attack.t1110.003
    - attack.initial-access
    - attack.t1078.004
logsource:
    product: azure
    service: signinlogs
detection:
    selection:
        ResultType:
            - 50126
            - 50053
            - 50034
            - 50057
    condition: selection
falsepositives:
    - Users mistyping passwords after a forced reset
level: low
---
title: Entra ID Password Spray - 10+ Distinct Accounts From One IP in 5 Minutes (ARTEX / Storm-0940)
id: 6001a000-0001-4000-8000-000000000002
status: experimental
description: One source IP failing sign-in against 10 or more distinct accounts within 5 minutes, consistent with ARTEX-automated cross-tenant password spraying attributed by Microsoft to Storm-0940 (TL-2026-6001).
references:
    - https://therecord.media/artex-ai-tool-fbi-flash-alert
author: Threadlinqs Team
date: 2026-11-13
tags:
    - attack.credential-access
    - attack.t1110.003
    - attack.initial-access
    - attack.t1078.004
correlation:
    type: value_count
    rules:
        - tl_6001_entra_failed_signin
    group-by:
        - IPAddress
    timespan: 5m
    condition:
        gte: 10
        field: UserPrincipalName
falsepositives:
    - Corporate NAT/VPN egress after a mass password expiry or SSO outage
    - Authorized password-spray assessments by internal red team
level: high
```

**Known false positives:**

- Corporate NAT or VPN egress IP producing many failures after a mass password expiry or SSO outage
- Authorized internal red-team password-spray exercises
- Misconfigured service or mail clients retrying stale credentials for many mailboxes

**Tuning notes:** Tune distinct_users (10) and failures (20) per 5-minute bin to tenant baseline; allow-list corporate egress IPs and registered red-team infrastructure; raise severity when scripted_ua=yes or the IP also appears in TL-2026-6001 network IOCs. Ensure the index/sourcetype names match your Splunk Microsoft Cloud Services inputs.

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- Langflow CVE-2025-3248 Code-Validation Exploitation and Cloud-Metadata SSRF Probing (locked: unlock with TL Creds)
- ARTEX Agent Host Artifacts - artx Temp Paths, Intent-ID Scripts and MARKER123/ARTX123 Commands (locked: unlock with TL Creds)
- ARTEX Console Listener and Non-Browser LLM API Egress from Hosts (locked: unlock with TL Creds)
- Scripted-User-Agent Successful Entra ID Sign-ins (ARTEX Valid Cloud Account / Token Abuse) (locked: unlock with TL Creds)
- ARTEX Container Image Pull or Run - Untrusted and Trojanized Image Risk (Troj/Bckdr-SDA) (locked: unlock with TL Creds)
- Web Proxy Access to [indicator] LLM Reseller or Exposed ARTEX Open Directory ([indicator]) (locked: unlock with TL Creds)
- Entra ID Service Principal Credential Addition / OAuth Grant Burst (ARTEX Privilege Escalation) (locked: unlock with TL Creds)
- ARTEX Infrastructure Egress - Reported Proxy/C2 IPs, [indicator] LLM Reseller and Console Ports (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sample indicators (3 of 48)

Defanged, TLP:CLEAR.

- IP address: `38.244.50[.]120`
- Domain: `xcai[.]pro`
- IP address: `101.53.80[.]20`

Full set of 48 indicators unlocks with TL Creds: https://threadlinqs.com/tl-creds

## Sources

- [FBI flash alert says Artex tool automating cyberattacks at 'machine speed' (The Record)](https://therecord.media/artex-ai-tool-fbi-flash-alert)
- [ARTEX: Agentic Asbestos (Censys)](https://censys.com/blog/artex-agentic-asbestos/)
- [Amazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents (AWS Security Blog)](https://aws.amazon.com/blogs/security/amazon-threat-intelligence-observes-threat-actors-running-cybercrime-campaigns-with-artex-ai-agents/)
- [Unknown threat actor uses ARTEX to target South Korean finance (CrowdStrike)](https://www.crowdstrike.com/en-us/blog/unknown-threat-actor-uses-artex-to-target-south-korean-finance/)
- [Microsoft links ARTEX-assisted Storm-0940 activity to ConnectWise compromise (Silicon UK)](https://www.silicon.co.uk/cybersecurity/microsoft-connectwise-artex-631988)
- [ARTEX attackers apply agentic AI (Sophos)](https://www.sophos.com/en-us/blog/artex-attackers-apply-agentic-ai)
- [ARTEX Cyberattacks: AI Agents Steal Credentials and Signing Certificates (eSecurity Planet, citing Check Point Research)](https://www.esecurityplanet.com/threats/news-artex-ai-agents-credential-theft/)
- [ARTEX AI cyberattacks on Japanese companies (eSecurity Planet, citing GMO Cybersecurity by Ierae)](https://www.esecurityplanet.com/cybersecurity-threats/news-apac-artex-ai-cyberattacks-japan-companies/)
- [ConnectWise suspends cloud backup service after supply chain attack (The Record)](https://therecord.media/connectwise-suspends-cloud-backup-service-supply-chain-attack)
- [Autumn-27 GitHub account (ARTEX developer; repository removed)](https://github.com/Autumn-27)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-6001
