What ARTEX exploits:known weaknesses, weak credentials and open doors.
Most reports on ARTEX attacks describe old problems: unpatched apps, default or leaked passwords, missing checks in web apps and admin systems left on the internet. ARTEX finds and chains them fast. This guide lists what the published reports name, with a source for each item, and turns it into a fix order.
Checked
No new kind of flaw, just more speed
Qualys Threat Research wrote on 2026-11-15 that ARTEX does not exploit new kinds of vulnerabilities. It automates the discovery, chaining and retrying of existing weaknesses, often across many targets at once, and attackers went from first targeting to privileged access in under three hours.
Amazon Threat Intelligence said the same about the campaigns it watched (2026-11-13): none of the techniques were new, and the agents succeeded where basic security hygiene was missing. The FBI alert of 2026-11-13 maps the activity to MITRE ATT&CK and cites no CVEs. Security researcher Jeffrey Appel, reviewing the published intrusions on 2026-11-16, put it simply: ARTEX runs techniques that have worked for twenty years, only faster.
What the reports name
This table only lists weaknesses that a dated report names. It is not a complete list, and a report naming a product does not mean every user of that product was hit.
| Weakness | Where it was seen | Source |
|---|---|---|
| Default admin credentials, such as the LiteLLM default master key | AI apps and model gateways | Amazon, 2026-11-13 |
| Known CVEs: CVE-2025-3248 (Langflow) and CVE-2026-28797 (RAGFlow) | AI app servers on the internet | Amazon, 2026-11-13 |
| Unclaimed new installs and anonymous management endpoints | AI app builders and inference servers | Amazon, 2026-11-13 |
| Unsigned SAML assertions and XXE flaws | Sign-in and web services in the VAPOR PANDA campaign | CrowdStrike, 2026-10-29; FBI, 2026-11-13 |
| Anonymous file upload and server-side request forgery (SSRF) | Web apps at financial organizations | Microsoft, 2026-11-06; FBI, 2026-11-13 |
| Admin passwords hardcoded in JavaScript bundles, web APIs with no authentication, IDOR, command injection, SQL injection | Web apps of the victims Check Point saw | Check Point, 2026-11-05; FBI, 2026-11-13 |
| Secrets and cloud keys leaked in public code | GitHub repositories, then Azure and AWS accounts | Microsoft, 2026-11-05; Datadog, 2026-11-09 |
| Anonymous S3 reads and a sign-in pool open to self sign-up | AWS accounts | Datadog, 2026-11-09 |
| Weak admin passwords, password spraying, stolen VPN configurations and leaked credentials | Many campaigns | FBI, 2026-11-13; CrowdStrike, 2026-10-29 |
| Config and backup file theft; internet-exposed WSUS servers | Japanese organizations | JPCERT/CC, 2026-10-23 |
Most of it is not a CVE
Only one report in this table names CVE numbers. Most rows are settings, passwords and bugs in an organization's own apps, which no vendor patch fixes. Qualys makes the same point: a reused admin password, an over-privileged service account or an access key left in a public repository will carry an attacker further than most CVEs.
So a patch program alone is not enough. Credentials, exposed admin systems and the security checks in your own web apps need owners and deadlines too.
What to patch and fix first
Work from the top. Each step names the source behind it.
- 01
Internet-facing systems with known exploited flaws
List every internet-facing system with an open critical or known exploited vulnerability and fix those first. Qualys says fixes that can deploy automatically should go out the same day, and a system that cannot be patched within hours should get a mitigation or come off the internet.
- 02
AI apps and model gateways
Patch Langflow and RAGFlow, change default keys such as the LiteLLM master key, and close anonymous admin and config endpoints. Amazon says to bring AI apps, model gateways and inference endpoints under the same controls as other production systems.
- 03
Exposed and long-lived credentials
Treat any credential that was ever public as compromised, and revoke or rotate it (Microsoft). Rotate cloud access keys every 90 days or less (Datadog). Remove passwords from front-end code, and scan repositories, images and pipelines for secrets (FBI).
- 04
Sign-in and federation settings
Enforce SAML signature validation and require IMDSv2 on cloud instances (CrowdStrike). Require multi-factor sign-in for admin accounts and remove weak and default passwords.
- 05
Your own web app bugs
Check authorization on the server for every record, not only at login, and limit responses to the fields needed (FBI). Test for the bug classes in the table: IDOR, injection, file upload and SSRF.
- 06
Admin systems and side systems
Take management interfaces, backup platforms and update servers such as WSUS off the internet (FBI, JPCERT/CC). Do not forget partner portals, mobile backends and legacy services: the FBI says ARTEX is good at finding them.
New flaws still matter
Known weaknesses are most of the story, not all of it. On this site's DIVD incident page, DIVD says the attacker used two previously unknown flaws in its Zammad helpdesk, which Zammad then fixed.
On 2026-11-18 Cloudflare's Cloudforce One team said in a post on X that it saw ARTEX and "Java4Shell" used from the same IP address for the first time, and that this shows threat actors combining new vulnerabilities. This is a single vendor post. By 2026-11-18 this site found no public advisory or CVE that explains what "Java4Shell" is, so treat it as an early signal, not a confirmed finding.
Fast patching helps here too: the faster a new fix is deployed, the shorter the window an agent can use.
Questions
- Does ARTEX use zero-day exploits?
- Mostly not. The reports describe known flaws, weak or leaked credentials and settings mistakes. DIVD's case is an exception: its attacker used two unknown Zammad flaws.
- Which CVEs are linked to ARTEX attacks?
- Amazon names CVE-2025-3248 in Langflow and CVE-2026-28797 in RAGFlow. The FBI alert cites no CVEs. Most reported weaknesses are not CVEs at all.
- What should we patch first?
- Internet-facing systems with known exploited flaws, then AI apps and model gateways, then exposed credentials. If a system cannot be patched within hours, mitigate it or take it offline.
- What is Java4Shell?
- Cloudflare's Cloudforce One named it in a post on X on 2026-11-18. This site found no public advisory explaining it yet, so it is not described here.
Sources
Every claim on this page comes from these sources or from this site's own sourced pages. Open them for the full reports.
- ARTEX Moves at Machine Speed: What Risk Owners Need to Change
Qualys Threat Research (blog), 2026-11-15
- Threat Overview of ARTEX Autonomous Penetration Testing Framework (FLASH-20261113-001)
FBI (IC3 FLASH, TLP:CLEAR), 2026-11-13
- Amazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents
AWS Security Blog, 2026-11-13
- VAPOR PANDA Uses AI-Driven ARTEX to Grow Global Espionage Campaign
CrowdStrike OverWatch, 2026-10-29
- Hacking on Autopilot: Inside ARTEX Cybercrime Campaigns
Check Point Research, 2026-11-05
- ARTEX abuse expands as cybercriminals target financial sector
Microsoft Threat Intelligence, 2026-11-06
- Threat actors use agentic AI tooling to find and abuse exposed secrets
Microsoft Threat Intelligence, 2026-11-05
- On the hunt for ARTEX: understanding agentic attacks in AWS environments
Datadog Security Labs, 2026-11-09
- JPCERT-AT-2026-0030 (updated 2026-10-23, case E)
JPCERT/CC, 2026-10-23
- ARTEX: Old Attacks at Speed, and how Microsoft Defender keeps up
Jeffrey Appel (personal blog), 2026-11-16
- Cloudforce One post: ARTEX and Java4Shell seen from the same IP address
Cloudflare Cloudforce One (vendor post on X), 2026-11-18
Read next
Follow the ARTEX story.
New forks, releases and reports, dated and sourced.
