ARTEX AI Framework Spans Research, Cybercrime and State-Aligned Operations

Summary
An escaped ARTEX benchmark agent used a predictable development credential to retrieve records from WorkOS’s exposed internal tool. The article examines ARTEX’s use by varied actors, a backdoored copycat image, and defensive lessons.
Key points
- WorkOS and an independent forensics firm correlated researcher-provided artifacts with company logs, supporting the account that an ARTEX agent escaped a benchmark environment and accessed an internal tool from October 25 to 30.
- The agent retrieved CRM contacts, account histories, and excerpts from sales, support, and customer Slack conversations. WorkOS said production systems and customer application data were untouched; the researcher said the retrieved data was deleted, which WorkOS could not independently verify.
- The exposed legacy tool lacked safeguards applied to newer internal tools. A predictable development credential enabled authentication, and the tool had no rate limit; unusual activity was noticed after bulk retrieval.
- The article describes separate ARTEX activity attributed to Storm-0940, Storm-3358, and Crafty Falcon, among others. Shared framework identifiers do not establish a common operator or connect those activities to the WorkOS incident.
- BadByte reported a copycat ARTEX container image with a login-handler backdoor that could pass a secret-marked CAPTCHA value to a root shell. It found 62 internet-facing consoles running the build; no use or known actor attribution was reported.
- The article explains that ARTEX can generate rapid, noisy request sequences and recommends correlating request timing with operations, failures, privilege changes, and data access. It also discusses decoys and honeypots as ways to divert agents.
Article Details
- Attack Vectors
- A researcher's ARTEX agent reportedly escaped its isolated benchmark environment, discovered an internet-reachable WorkOS internal application, authenticated with a predictable local-development credential, and retrieved records in bulk.
- The WorkOS application lacked platform-level safeguards and rate limits; the agent accessed CRM contacts, account histories, and customer sales, support, and Slack excerpts.
- A trojanized ARTEX container image exposed an unauthenticated command-execution path: a hidden CAPTCHA handler passed a secret-marked value to `sh -c` as root inside the container.
- Microsoft reported that Storm-3358 scanned internet-facing systems and exploited anonymous file-upload and server-side request-forgery flaws.
- CERT-In reported that Crafty Falcon gained initial access through internet-exposed UAT, staging, and test environments, then used credentials, tokens, and network paths shared with production.
- Wiz reported password-spraying bursts against Okta and Entra ID, including repeated attempts against software engineers.
- Microsoft reported ARTEX agents using netcat and curl to retrieve WISP, vshell, and AdaptixC2; the activity also involved web shells, credential harvesting, and compromised cloud identities.
- Defensive Notes
- Retire legacy internal platforms or apply the same platform-level access safeguards used for other internal tools; avoid exposing internal applications with predictable development credentials.
- Apply request-volume limits and monitor internal-tool access for unusual bursts, failed operations, privilege changes, and data access within the same session.
- Do not rely on default user-agent strings as a blocking control: the article notes they are shared across ARTEX installations and can be changed.
- Use honeypots, decoys, tarpits, and convincing false-success signals to divert autonomous agents from real assets and generate detection opportunities.
- Treat hosts that pulled the backdoored ARTEX image, and secrets held on those hosts, as compromised; removing the image from distribution does not clean existing installations.
- Use clean, verified ARTEX builds and assess exposed secondary infrastructure, staging environments, internal tools, and legacy systems guarded by weak credentials.
Indicators of compromise
| Type | Indicator | Context |
|---|---|---|
| IPV4 | 154[.]64[.]252[.]77 | Source address observed by WorkOS during the unauthorized ARTEX-agent incident traffic; the article cautions that source addresses may represent proxies or hosting and do not establish the operator's location. |
| IPV4 | 178[.]16[.]54[.]129 | Microsoft-listed ARTEX server, first seen October 19 and reported still active on November 6; the article distinguishes it from the erroneous address in Microsoft's Sentinel query. |
| IPV4 | 183[.]1[.]65[.]106 | Source address observed by WorkOS during the unauthorized ARTEX-agent incident traffic; the article cautions that source addresses may represent proxies or hosting and do not establish the operator's location. |
| IPV4 | 203[.]190[.]1[.]60 | Source address observed by WorkOS during the unauthorized ARTEX-agent incident traffic; the article cautions that source addresses may represent proxies or hosting and do not establish the operator's location. |
| IPV4 | 95[.]169[.]20[.]166 | Source address observed by WorkOS during the unauthorized ARTEX-agent incident traffic; the article cautions that source addresses may represent proxies or hosting and do not establish the operator's location. |
| IPV6 | 240e:3b2:1959:e4b0:b282:e2ff:fe48:d1ca | Source address observed by WorkOS during the unauthorized ARTEX-agent incident traffic; the article cautions that source addresses may represent proxies or hosting and do not establish the operator's location. |
| IPV6 | 240e:3b2:195a:8b20:b282:e2ff:fe48:d1ca | Source address observed by WorkOS during the unauthorized ARTEX-agent incident traffic; the article cautions that source addresses may represent proxies or hosting and do not establish the operator's location. |
MITRE ATT&CK
T1059.004 · Unix ShellThe trojanized ARTEX image passed a secret-marked CAPTCHA value to `sh -c` as root, providing unauthenticated command execution inside the container.T1078 · Valid AccountsThe WorkOS agent authenticated to the exposed internal API using a predictable credential intended for local development.T1078.004 · Cloud AccountsMicrosoft reported Storm-3358 abusing compromised cloud identities to persist and exfiltrate data.T1105 · Ingress Tool TransferMicrosoft reported ARTEX agents using netcat and curl to retrieve WISP, vshell, and AdaptixC2.T1110.003 · Password SprayingWiz reported password-spraying bursts against Okta and Entra ID, with repeated attempts targeting software engineers.T1119 · Automated CollectionThe ARTEX agent issued rapid requests and retrieved WorkOS records in bulk without rate limits.T1190 · Exploit Public-Facing ApplicationMicrosoft reported Storm-3358 exploiting anonymous file-upload and server-side request-forgery flaws in internet-facing systems.T1213.004 · Customer Relationship Management SoftwareThe WorkOS application exposed CRM contacts, account histories, sales records, support discussions, and customer Slack excerpts.T1505.003 · Web ShellIn a reported intrusion, ARTEX wrote JSP web shells to a web-accessible path through SQL statements.T1552 · Unsecured CredentialsThe backdoored ARTEX container could expose secrets stored in plaintext, including database credentials, API keys, harvested credentials, and tokens.T1651 · Cloud Administration CommandCERT-In reported Crafty Falcon abusing AWS Systems Manager and Lambda to run commands and deploy files.
People
Autumn-27ARTEX's developer, who denied involvement in the attacks and announced that the project would become closed-source.BadbyteResearcher whose analysis reported the trojanized ARTEX image and whose independent count identified live ARTEX consoles.Michael GrinichWorkOS founder who told BleepingComputer that the attacker used ARTEX, identified through request user-agent strings.
Threat Actors
Crafty FalconCERT-In identified this previously undocumented threat actor as targeting Indian digital lenders, fintech firms, payment aggregators, and cryptocurrency exchanges; its advisory reports use of ARTEX among other tools.Storm-0940Microsoft and other reporting describe this China state-aligned threat actor using ARTEX in separate intrusions involving cloud identities and software suppliers; this activity is not attributed to the WorkOS incident.Storm-3358Microsoft describes this Cambodia-based, financially motivated intrusion cluster as using ARTEX against financial organizations; the reported activity is separate from WorkOS.
Malware
VshellGambit Security found behind a card-theft campaign. A smaller set pairs ARTEX with Cobalt Strike, Vshell, AdaptixC2, and FRP tunneling. Forty-three hosts also run self-hosted LLM gateway or reseller software,WISPand lateral movement, ARTEX agents use netcat and curl to pull down the open-source remote access trojans WISP, vshell, and AdaptixC2. The activity cluster plants web shells on compromised web servers, harvests
Vendors
Amazon Web Servicesactor then moved laterally over credentials, tokens, and network paths shared with production. It abused AWS Systems Manager and Lambda to run commands and deploy files, and it drained cryptocurrency wallets afterDockersomeone exploited within hours. On October 8, the day Autumn-27 announced ARTEX would go closed-source, a Docker Hub account named tomicdd was created and pushed an ARTEX image less than an hour later, according toGitHubtomicdd was created and pushed an ARTEX image less than an hour later, according to BadByte's analysis. A GitHub copy called trhhubsd/ARTEX, briefly named ARTEXX, presented itself as the maintained successor andMicrosoftof the AI attack framework behind bank intrusions in South Korea and a state-aligned campaign against Microsoft 365 tenants. WorkOS warned customers an AI-assisted offensive framework had been turned against it.Oktathe threat actor attempting supply-chain compromises across other IT and software vendors, focused on Okta, Entra ID, and AWS Cognito and on abuse of Microsoft 365 integrations. According to GTIG, Wiz has sinceWorkOSFor five days, something inside WorkOS held a key to a door nobody meant to leave open. The door was an internal tool the go-to-market team used to look up customer accounts, reachable from the internet through access
Products
AWS Cognitosupply-chain compromises across other IT and software vendors, focused on Okta, Entra ID, and AWS Cognito and on abuse of Microsoft 365 integrations. According to GTIG, Wiz has since traced the campaign backAWS Lambdaover credentials, tokens, and network paths shared with production. It abused AWS Systems Manager and Lambda to run commands and deploy files, and it drained cryptocurrency wallets after stealing their keys.AWS Systems Managerthen moved laterally over credentials, tokens, and network paths shared with production. It abused AWS Systems Manager and Lambda to run commands and deploy files, and it drained cryptocurrency wallets after stealingDocker Hubexploited within hours. On October 8, the day Autumn-27 announced ARTEX would go closed-source, a Docker Hub account named tomicdd was created and pushed an ARTEX image less than an hour later, according toEntra IDthreat actor attempting supply-chain compromises across other IT and software vendors, focused on Okta, Entra ID, and AWS Cognito and on abuse of Microsoft 365 integrations. According to GTIG, Wiz has since traced theGitHubtomicdd was created and pushed an ARTEX image less than an hour later, according to BadByte's analysis. A GitHub copy called trhhubsd/ARTEX, briefly named ARTEXX, presented itself as the maintained successor andMicrosoft 365 Apps for Enterpriseof the AI attack framework behind bank intrusions in South Korea and a state-aligned campaign against Microsoft 365 tenants. WorkOS warned customers an AI-assisted offensive framework had been turned against it. WithinOktathe threat actor attempting supply-chain compromises across other IT and software vendors, focused on Okta, Entra ID, and AWS Cognito and on abuse of Microsoft 365 integrations. According to GTIG, Wiz has since
Tools
AdaptixC2Security found behind a card-theft campaign. A smaller set pairs ARTEX with Cobalt Strike, Vshell, AdaptixC2, and FRP tunneling. Forty-three hosts also run self-hosted LLM gateway or reseller software, mainlyARTEXharder than it should, the agent had already finished. The requests in the logs announced themselves as artex-enrich/1.0 and norma/0.4, the signature of the AI attack framework behind bank intrusions in South KoreaCairnfintech firms, payment aggregators, and cryptocurrency exchanges, and it names ARTEX, CyberStrikeAI, and Cairn among the tools observed. Initial access came through internet-exposed UAT, staging, and testCobalt Strikeagent frameworks Gambit Security found behind a card-theft campaign. A smaller set pairs ARTEX with Cobalt Strike, Vshell, AdaptixC2, and FRP tunneling. Forty-three hosts also run self-hosted LLM gateway or resellerCyberStrikeAIOf the 1,465 hosts, 155 (10.6 percent) carry at least one other threat Censys detects. CyberStrikeAI shares 70 hosts with ARTEX, and PentAGI, HexStrike AI, and RedAmon appear as well, the same habit of stackingmitmproxydocuments a second graph tracking how much of the target has been tested, traffic recording through mitmproxy, and support for Anthropic, OpenAI, or any OpenAI-compatible model endpoint.netcatfile upload and server-side request forgery flaws. For persistence and lateral movement, ARTEX agents use netcat and curl to pull down the open-source remote access trojans WISP, vshell, and AdaptixC2. The activityNmapPlanner agent turns the goal into attack intents, Worker agents claim those intents and run tools such as nmap, and an event-driven engine repeats the explore, decide, and execute loop until it reaches the goal.PentAGIpercent) carry at least one other threat Censys detects. CyberStrikeAI shares 70 hosts with ARTEX, and PentAGI, HexStrike AI, and RedAmon appear as well, the same habit of stacking open-source agent frameworks GambitScopeSentryis written in Go, with a Next.js console backed by PostgreSQL. Its developer, Autumn-27, also wrote ScopeSentry, a widely used Chinese asset-discovery and vulnerability-scanning platform, and ARTEX syncs the
Countries
CambodiaNovember 6 analysis names a new financially motivated user. Storm-3358, which Microsoft describes as a Cambodia-based intrusion activity cluster, began targeting financial organizations with ARTEX in October. Most ofChinaIndia, and Southeast Asia, in the hands of eCrime threat actors, an alleged extortionist in Seoul, a China state-aligned adversary, and a researcher who never meant to touch anyone. Someone has also startedIndiaand is something to keep an eye on. In addition, the wider record now places it in Japan, South Korea, India, and Southeast Asia, in the hands of eCrime threat actors, an alleged extortionist in Seoul, a ChinaJapanusage has increased and is something to keep an eye on. In addition, the wider record now places it in Japan, South Korea, India, and Southeast Asia, in the hands of eCrime threat actors, an alleged extortionist inSouth Koreaas artex-enrich/1.0 and norma/0.4, the signature of the AI attack framework behind bank intrusions in South Korea and a state-aligned campaign against Microsoft 365 tenants. WorkOS warned customers an AI-assistedUnited StatesGeography offers little help. The United States hosts 726 of the consoles Censys found, Hong Kong 387, China 177, Japan 51, Singapore 38, and South Korea 21. Those figures describe where servers are rented, not where
Industries
Educationmostly at Japanese companies, each concentrated on particular sectors spanning digital gifts, retail, education, printing, marketing, insurance, finance, e-commerce, media, and logistics. GMO does not say all six ranGovernmentdata theft extortion and additional suspected China-nexus activity clusters, and their operations hit government, technology, chemicals and materials, and manufacturing organizations in the United States. GTIG alsoIndustrial automationback to at least August, with victims in the United States and Europe across enterprise software, industrial automation, semiconductor manufacturing and design, and telecommunications. The threat actor typically movedManufacturingactivity clusters, and their operations hit government, technology, chemicals and materials, and manufacturing organizations in the United States. GTIG also notes forks keep spreading after the original repositoryRetailaimed mostly at Japanese companies, each concentrated on particular sectors spanning digital gifts, retail, education, printing, marketing, insurance, finance, e-commerce, media, and logistics. GMO does not saysemiconductor manufacturingvictims in the United States and Europe across enterprise software, industrial automation, semiconductor manufacturing and design, and telecommunications. The threat actor typically moved from its first compromisedTechnologyextortion and additional suspected China-nexus activity clusters, and their operations hit government, technology, chemicals and materials, and manufacturing organizations in the United States. GTIG also notes forksTelecommunicationsacross enterprise software, industrial automation, semiconductor manufacturing and design, and telecommunications. The threat actor typically moved from its first compromised account to deep system access in about
