The Federal Bureau of Investigation has issued a flash alert warning that an open-source artificial intelligence framework known as Artex is being leveraged by cybercriminals and nation-state actors to conduct automated attacks at unprecedented speeds. According to the advisory, the tool enables threat actors to execute multiple stages of a cyber intrusion, ranging from initial reconnaissance to final data exfiltration, with minimal human intervention. The FBI reports that the framework has significantly compressed the timeline for successful breaches, with some operations moving from target identification to the acquisition of privileged access in less than three hours.
Artex was originally published on GitHub in August by a user operating under the handle Autumn-27. The developer, identified as Chinese cybersecurity engineer Li Puhua, issued a public statement on October 8 acknowledging that the software had been co-opted for malicious purposes. While Li Puhua denied personal responsibility for these criminal activities and subsequently deleted the repository, the FBI confirms that the framework remains widely available. It continues to circulate through community-maintained forks, archived versions, localized variations, and even a commercial iteration that has emerged in the wake of the original repository's removal.
The framework functions by translating high-level tactical objectives provided by a user into actionable machine commands carried out by autonomous AI agents. A notable feature of Artex is its ability to integrate with a wide variety of large language models, including frontier models such as OpenAI’s GPT and Anthropic’s Claude, as well as open-source or China-based alternatives like DeepSeek, GLM, Grok, and Kimi. Research cited by the FBI indicates that Chinese-origin models are the most frequently utilized, with the DeepSeek Flash family being particularly prevalent due to its cost-effectiveness and perceived lack of restrictive guardrails when configured for penetration testing tasks.
Censys research referenced in the alert identified more than 1,600 publicly exposed Artex console instances, though the actual number of deployments is likely higher, as many systems remain hidden behind firewalls or within private networks. The tool has been implicated in a broad range of sectors, including government, healthcare, financial services, telecommunications, defense, and semiconductor manufacturing. Recent incidents include high-profile attacks on financial institutions in Japan and South Korea, as well as a supply chain compromise involving IT management software provider ConnectWise, which Microsoft attributed to Chinese state-backed actors.
Amazon has corroborated these findings, detailing campaigns where disparate threat groups utilized Artex to steal credentials. These groups reportedly organized their operations into self-sustaining cycles, using stolen API keys to fund the AI inference costs required to power further credential-harvesting efforts. Additionally, Amazon observed the framework being used to facilitate the bypass of know-your-customer verification processes at cryptocurrency exchanges through the deployment of real-time deepfake injections.
The FBI notes that Artex is particularly adept at mapping trust relationships between organizations to identify and exploit less-secure entry points. These targets often include third-party vendor portals, mobile backends, legacy services, and internal applications that may receive less rigorous security oversight than core infrastructure. By automating these complex, multi-target operations, the framework allows even threat actors with limited technical expertise to scale their activities significantly.
While the high volume of attempts generated by Artex agents is likely to trigger security alerts, the speed at which the tool operates creates a substantial challenge for human defenders. The FBI warns that the framework’s ability to conduct concurrent intrusion campaigns across multiple targets simultaneously represents a fundamental shift in the velocity and scale of modern cyber threats. Organizations are advised to monitor for the presence of these automated agents and to prioritize the hardening of third-party integrations and auxiliary business systems that have become primary targets for the tool.





