FBI warns ARTEX attacks leave cloud defenders hours to act

FBI warns ARTEX attacks leave cloud defenders hours to act

Add CyberWorldOpsto your preferred sources on Google
FBI warns ARTEX attacks leave cloud defenders hours to act
AI

Illustrative image generated with AI

Some attackers using ARTEX, an AI-driven security-testing framework, have reached privileged access in less than three hours, the FBI warned in a Friday advisory reported by BleepingComputer on November 15. The warning includes the October ConnectWise breach, where compromised application identities gave intruders a route into customers’ Microsoft 365 environments through permissions those customers had already granted.

The exposure warrants attention from organizations running internet-facing applications and cloud integrations, especially Microsoft 365 customers that have authorized the affected ConnectWise applications—even if they later stopped using them. BleepingComputer reported that the FBI has tracked abuse of ARTEX by multiple threat actors since at least August 2026, spanning finance, government, technology, defense, healthcare and other sectors.

Existing application permissions opened customer environments

In its October 21 investigation, Microsoft assessed that China-state-aligned actor Storm-0940 carried out the ConnectWise supply-chain compromise. Microsoft observed the actor using ARTEX from at least September to target hundreds of organizations worldwide, with several cloud intrusions reaching privileged access and multiple cases taking less than three hours from initial targeting.

The ConnectWise attack involved identities belonging to its Cloud Backup service, formerly SkyKick Cloud Backup. By compromising those identities, attackers could authenticate to customer tenants—separate Microsoft 365 environments—without an interactive user sign-in. ARTEX accelerated reconnaissance, password spraying, which tries passwords across multiple accounts, and abuse of OAuth applications, which hold permissions to access cloud resources.

BleepingComputer reported that attackers attempted authentication against more than 2,250 downstream tenants, while Huntress identified at least 162 compromised tenants. Those figures measure different outcomes: attempted authentication and identified compromise.

Microsoft saw subsequent activity through Microsoft Graph, the interface applications use to access Microsoft cloud resources. This included reconnaissance, changes to application credentials and attempts to alter privileges for broader or persistent access. Its observed-technique mapping also lists cloud-role additions, including self-assignment of global administrator through a ConnectWise management application, and bulk retrieval of secrets from Azure Key Vaults.

Microsoft said the campaign did not stem from a compromise or vulnerability in its own corporate systems, products or services. It also cautioned that finding an affected application or a successful attacker authentication does not necessarily mean customer data was accessed or stolen.

Storm-0940 has operated since 2021, has no known industry aliases, and historically has pursued data theft through password spraying and exploitation of internet-facing applications. Microsoft says it has also maintained cloud access by adding credentials to existing OAuth applications and used Microsoft Graph to retrieve email and other data. Its ARTEX targets span aerospace, defense, automotive, chemicals, energy, finance, government, legal services, manufacturing, media, military, semiconductors, software, technology, think tanks and nonprofits.

How ARTEX automates the attack sequence

ARTEX was built for penetration testing: assessing security by attempting attacks. Written in Go, it delegates tasks to AI agents that can survey targets, identify and exploit weaknesses, obtain greater privileges and steal data with limited human direction, BleepingComputer reported.

The operator supplies objectives; ARTEX turns them into commands for specialized agents. It relies on external large language models rather than its own model, with integrations including Anthropic Claude, OpenAI GPT, DeepSeek, GLM, Grok and Kimi.

The FBI says this approach lets people with limited technical expertise run complex operations against several targets at once. The bureau highlighted weaknesses in systems that can receive less attention, including partner portals, employee applications, older services and third-party integrations. Automation may produce conspicuous traffic, but an alert can leave defenders little time to interrupt the intrusion.

Developer Autumn-27 released ARTEX on GitHub in July 2026, then removed the repository on October 8 following reports of abuse. The FBI says copies remain accessible through community forks, archives, localized versions and the commercial ARTEX PRO edition. Researchers had found more than 1,600 publicly exposed ARTEX consoles by early November, with the total deployment count believed to be substantially higher.

Other investigations document broader use

BleepingComputer reported that the FBI advisory also covers a campaign against South Korean financial institutions that exposed personal information belonging to approximately 67,000 people. A Chinese-speaking actor paired ARTEX with Claude Code agents against banks including Shinhan Bank, KB Kookmin Bank and Hana Bank. CrowdStrike found exposed attacker infrastructure holding ARTEX configuration files, Claude Code session histories and other intrusion records.

In a separate case, South Korean police arrested seven suspects linked to attacks on more than 90 organizations. Authorities said approximately 8.2 million data records were stolen from hospitals, educational institutions and businesses; ARTEX reportedly featured in most of those intrusions.

More recently, CrowdStrike reported that Chinese cyberespionage group Vapor Panda used ARTEX against more than 200 organizations in 14 countries, compromising at least nine. Agents surveyed systems, exploited vulnerabilities, installed web shells—scripts providing remote access—and stole credentials. They also pursued alternative routes when an initial attempt failed.

Review application identities and preserve evidence

Microsoft advises administrators to keep affected ConnectWise connections disabled until ConnectWise supplies formal restoration instructions. Before deleting enterprise applications or making other irreversible changes, preserve their configuration and investigation evidence.

The review should cover every tenant where affected applications are or were installed or authorized. Search using application IDs rather than display names, and record each tenant’s service-principal Object ID, the identifier for its local application identity. Microsoft calls for inventorying all four application IDs in its profile; its hunting examples identify these three:

  • ConnectWise Cloud Backup: 506e08f9-6e9a-4da4-942b-a099a03644ae
  • ConnectWise - MS Teams: 24205314-2b89-4e07-a2a4-579e898a5640
  • ConnectWise - MS Graph: 0de14c77-b9b0-4137-9360-34878a929827

Preserve available Entra sign-in and audit records, Purview Audit records, security-monitoring data and any previously configured Graph telemetry. Microsoft says urgent containment should proceed without waiting for exports. Review activity from at least October 15, 2026, extending service-principal sign-in analysis to October 1 where records permit. Investigate earlier related activity, including authentication failures and abnormal volumes, and examine successful and failed sign-ins separately.

Microsoft’s Defender XDR and Sentinel queries help investigators establish which application identities authenticated, where requests originated and which resources they targeted, then examine subsequent Graph requests and responses. Other queries cover Azure Key Vault, Storage and Azure Resource Manager activity. Assess missing Graph results alongside application permissions and authentication evidence rather than treating an empty result as proof that access never occurred.

Microsoft lists 144.172.64[.]128, 144.172.64[.]99 and 144.172.64[.]60 for observed sign-ins and Graph activity. The VPN addresses 187.15.88[.]186, 187.15.90[.]217 and 187.15.89[.]186 have a narrower investigative context: October 15–16 activity involving compromised applications. Listed user agents are Python-urllib/3.12, python-requests/2.31.0, python-requests/2.34.2, curl/7.88.1 and azure-toolkit/1.0.

The FBI’s 13-page advisory provides campaign indicators, including IP addresses, attacker-control infrastructure and suspicious network activity. Researcher MalwareMustDie separately released an open-source detection pack on Friday covering network requests, identity abuse, cloud activity, malicious files and automated attack patterns. Its accompanying playbook ranks indicators by confidence and warns about stale infrastructure and false positives; the FBI likewise cautions that listed IP addresses may have changed.

Microsoft also promotes its Security Dashboard for AI as a way to combine security, identity and data-risk information from Defender, Entra and Purview to identify AI assets and prioritize security work.

For broader protection, the FBI recommends securing internet-facing applications, limiting access to cloud administration interfaces, checking application permissions and authentication logs, and applying request-rate limits to sensitive APIs. Defenders should watch for bursts of automated requests, unusual cloud activity, unexpected credential or privilege changes and large transfers of data. Organizations identifying suspected ARTEX activity should report it to the FBI or its Internet Crime Complaint Center.

Read next

Sources

Back to home

Latest Cybersecurity News

All cybersecurity news →