FBI FLASH says ARTEX AI framework automates intrusions at machine speed, with some reaching privileged access in under three hours
TLDR
- The FBI says malicious actors use ARTEX to automate intrusions, some reaching privileged access in under three hours.1
- ARTEX remains available through forks, archived copies and a commercial version after its developer removed the repository, the FBI says.1
- Amazon, Talos and SentinelOne document ARTEX campaigns including AI-model credential theft and China-nexus espionage.345
The FBI has released a FLASH, numbered FLASH-20261113-001 and dated 13 November 2026, on ARTEX, which it describes as an open-source, large language model-driven agentic platform that threat actors have weaponised to automate multiple phases of cyber intrusions at machine speed, from reconnaissance through data exfiltration.1 The Record reports that the FBI issued the alert on Friday.2 The FBI says a variety of malicious cyber actors have used ARTEX since at least August 2026 against organisations in financial services, government, technology, defence, semiconductor, energy, healthcare, manufacturing, telecommunications and legal sectors.1 It says some intrusions progressed from initial targeting to privileged access in under three hours, and that the framework lets actors with minimal technical expertise run complex, multi-target operations at the same time.1 The FBI names two notable campaigns, among others: financially motivated attacks on South Korean financial institutions that exfiltrated personal information from approximately 67,000 individuals, and a supply chain compromise of ConnectWise Cloud Backup customers that gave access to more than 2,250 downstream tenants.1
The FBI says the original developer, who published ARTEX on GitHub as Autumn-27, made the repository unavailable on 8 October 2026 after reports of abuse, but that the framework continues to spread through community forks, localised variations, archived copies and a commercial version called ARTEX PRO.1 The FBI and Talos date the original release to July 2026, while The Record says it was released in August.124 As of early November, the FBI says, researchers identified over 1,600 publicly exposed ARTEX console instances, with total deployments likely significantly higher; The Record attributes that figure to Censys.12 SentinelOne gives 1,608 live and publicly available consoles, and 1,465 exposed consoles on 3 and 4 November.5 Cloudflare says the ecosystem has grown since the 8 October takedown to over 12,000 repositories, with several hundred unique forks carrying lineage from the original, and that daily requests from AI penetration testing and exploitation tools have grown 30 times since 1 October.6 It says that as of 12 November, ARTEX and its forks make up over 60 per cent of the AI penetration testing orchestrators it knows of.6
The FBI says ARTEX has no model of its own and integrates with external providers including Anthropic Claude, OpenAI GPT, DeepSeek, GLM, Grok and Kimi, and that in attacker-controlled instances Chinese models are most prevalent, with DeepSeek Flash family models dominating.1 It says ARTEX excels at finding peripheral systems that get less security attention, including partner and supplier portals, mobile backends, employee applications, sales-support systems, legacy services, third-party integrations and applications built across business units.1 The FLASH lists initial access methods including unsigned SAML assertions, XXE flaws, anonymous file upload, SSRF, hardcoded administrator passwords in JavaScript bundles and missing API authentication, and says the high-volume, forensically noisy traffic may trigger alerts but the attack speed leaves limited time for a coordinated defender response before the objective is complete.1 Its recommendations include inventorying all internet-facing applications including those peripheral systems, restricting cloud administration and similar interfaces to trusted networks, applying least privilege to multi-tenant application permissions, rate-limiting portals that return customer records, scanning repositories, container images and pipelines for embedded secrets and rotating any that are exposed.1 For ConnectWise, it advises reviewing every Microsoft 365 tenant where affected ConnectWise applications are or were installed, searching by application ID, and reviewing activity from at least 1 October 2026.1
Other vendors published their own ARTEX findings on the same day. Amazon says its MadPot decoy sensors, set up to look like LLM inference APIs, captured requests from ARTEX agents between 3 October and 5 November that carried the agents' own records of their campaigns.3 Amazon assesses with moderate confidence that the campaigns used ARTEX and were run by disparate Chinese-nexus groups.3 The most prominent was a large-scale theft of credentials for AI models, in which stolen keys paid for the inference that powered further credential hunting; another targeted know-your-customer checks at two cryptocurrency exchanges with real-time deepfakes, and Amazon says the one vendor's liveness check whose protocol the agents reverse-engineered rejected the deepfake in every recorded attempt.3 Amazon says it cannot verify the outcomes the agents reported and that none of the techniques in these campaigns were new.3 Talos lists campaigns attributed by others: CrowdStrike linked an espionage campaign beginning 17 October to China-nexus actor VAPOR PANDA, which used ARTEX against more than 200 organisations in at least 14 countries with post-compromise activity confirmed at nine, and Microsoft attributed the ConnectWise activity to Storm-0940, which it assesses as China-state aligned.4 Talos adds that artifacts from the South Korean bank attacks support an association with an ARTEX environment but do not by themselves prove the complete framework was used in a successful attack.4 SentinelOne says each campaign it describes cost $145 to $1,500 in model time, and that in one suspected case an agent went from a leaked GitHub secret to Azure Key Vault data in five minutes.5 Cloudflare says its incident response team saw ARTEX use a compromised account-level CI/CD API token with no IP restrictions or rate limiting, creating Worker scripts with names such as artex-*, and that the evidence indicated the compromise most likely came through a third-party hyperscaler.6
Why it matters: the FBI says ARTEX favours overlooked partner portals, sales-support systems and third-party integrations and can reach privileged access in under three hours, so those peripheral systems need the same inventory and controls as flagship applications.1
Sources
- Threat Overview of ARTEX Autonomous Penetration Testing Framework (FLASH-20261113-001) (FBI)
- FBI flash alert says Artex tool automating cyberattacks at 'machine speed' (The Record)
- Amazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents (AWS Security Blog)
- ARTEX and beyond: AI-enabled penetration testing platforms (Cisco Talos)
- ARTEX Unmasked | The Rise of AI-Powered ARTEX in Cybercrime & Espionage (SentinelOne)
- Adapting to AI penetration testing orchestrators: The Modern Security Model (Cloudflare)