An open-source artificial intelligence framework named Artex is increasingly being weaponised by cybercriminals and nation-state threat actors to breach government agencies and dozens of critical infrastructure sectors worldwide, according to a flash alert issued by the US Federal Bureau of Investigation (FBI).
In an alert issued on Friday, the FBI warned that the platform can turn any large language model (LLM) into a powerful hacking tool capable of automating multiple stages of cyber intrusions at machine speed, from initial reconnaissance to data exfiltration, reports The Record.
The agency said Artex significantly reduces the time required to combine different attack techniques, allowing some breaches to escalate from initial targeting to privileged access in under three hours.
The framework also enables threat actors with limited technical expertise to conduct complex operations against multiple targets simultaneously, accelerating attacks and expanding their scale.
Proliferation despite developer repository removal
The tool was first published on GitHub in August by an account operating under the handle “Autumn-27”. Its developer, Chinese cybersecurity engineer Li Puhua, acknowledged in a GitHub statement on October 8 that malicious actors had abused the framework.
Li denied responsibility for any illegal activities carried out using the software and subsequently removed the GitHub repository.
However, the FBI said Artex continues to spread worldwide despite the deletion. Community forks, localised versions, archived copies and a commercial version have ensured that the framework remains available to threat actors.
Data from cybersecurity firm Censys showed more than 1,600 publicly exposed Artex console instances, while the actual number of global deployments is likely much higher when systems operating behind private networks or firewalls are taken into account.
Artex functions as a penetration-testing platform, translating plain-language tactical objectives into executable machine commands carried out autonomously by AI agents. It requires minimal human involvement beyond the initial instructions.
Chinese LLMs and DeepSeek Flash dominate usage
The framework can integrate a wide range of AI models, including leading platforms such as Anthropic Claude and OpenAI GPT, as well as open-source and Chinese-origin models, including DeepSeek, GLM, Grok and Kimi.
The FBI said an analysis of attacker-controlled Artex instances found that Chinese models were the most widely used. The DeepSeek Flash family dominated usage because of its technical capabilities, relatively low cost and lack of observable guardrails when configured for penetration testing.
Artex has been used for reconnaissance and data exfiltration across several sectors, including financial services, government, technology, defence, semiconductors, energy, healthcare, manufacturing, telecommunications and legal services.
The framework was also implicated in several recent cyber attacks targeting financial institutions in South Korea and Japan, triggering national scandals in both countries.
Tech giants uncover state-backed
Technology giant Microsoft said Chinese state-backed actors had used Artex in a recent supply chain intrusion targeting IT management software provider ConnectWise.
Meanwhile, Amazon published an independent study on Friday revealing that “disparate Chinese-nexus threat groups” were using Artex for large-scale credential theft to gain access to AI models.
According to the study, the attackers had developed a self-sustaining operation in which stolen keys were used to pay for the LLM inference that powered further credential-hunting activities.
Amazon Chief Information Security Officer (CISO) CJ Moses also highlighted a campaign targeting cryptocurrency exchanges, in which attackers attempted to bypass know-your-customer (KYC) verification through real-time deepfake injection.
Moses warned that the investigation showed how tools such as Artex “have progressed from experimental use to sustained, multi-objective offensive operations that run with minimal human intervention”.
Amazon further cautioned that by automating time-consuming exploitation stages, Artex allows smaller threat groups to launch multiple intrusion campaigns simultaneously on a scale that was previously beyond their reach.
Exploiting weak third-party security links
According to the FBI’s technical analysis, AI agents deployed through Artex can map trust relationships between organisations and compromise vulnerable third-party vendors.
The platform is particularly effective at identifying and exploiting systems that receive less security attention. These include partner and supplier portals, mobile backends, employee applications, sales-support systems, legacy services, third-party integrations and applications developed across different business units.
Although the framework generates large volumes of intrusion attempts that trigger security sensors, its operating speed makes it difficult for defenders to keep pace.
