
Top-Secured cyber watch: independent analysis based on the sources listed below. Top-Secured is not affiliated with the companies mentioned and will never ask you for your bank details or passwords.
On 13 November 2026, the FBI issued a FLASH alert on ARTEX, an open-source framework whose AI agents automate intrusions, and Amazon detailed a campaign in which these agents went after self-hosted AI gateways and platforms such as LiteLLM, Dify or Ollama. The weaknesses exploited are mostly basic: default credentials, open admin interfaces, leaked secrets.
What happened
On 13 November 2026, the FBI published a FLASH alert (FLASH-20261113-001) on ARTEX, an open-source penetration testing framework written in Go in which AI agents driven by large language models chain the steps of an intrusion with little human input. According to the FBI, criminal groups have used it since at least August 2026 against many sectors, and some intrusions went from first targeting to privileged access in under three hours. The developer removed the GitHub repository on 8 October 2026, but forks, mirrors and a paid version keep circulating; researchers counted more than 1,600 ARTEX consoles exposed online in early November.
The same day, Amazon described what its MadPot decoy sensors, built to look like AI model APIs, captured between 3 October and 5 November 2026: requests from ARTEX agents still carrying their working memory. Amazon attributes the activity, with moderate confidence, to several China-linked groups, and says it could not verify the results the agents reported.
What the attackers go after
In the first campaign described by Amazon, the agents hunted API keys for AI models, to use them at the victims’ expense, then moved on to cloud accounts. Targets included self-hosted LLM gateways (LiteLLM, One API, New API…) and AI application platforms (Dify, Langflow, RAGFlow, Open WebUI, Ollama…). The weaknesses were mostly basic: default administrator credentials such as the LiteLLM default master key, installs never initialised, management endpoints without authentication, SSRF to cloud metadata services, secrets left in JavaScript files or environment files, and known flaws such as CVE-2025-3248 (Langflow) and CVE-2026-28797 (RAGFlow). The FBI lists similar entry points, plus leaked credentials and stolen VPN configurations, followed by PHP or JSP web shells and added SSH keys.
What to do now
- List your AI services reachable from the internet (LiteLLM, Ollama, Open WebUI, Dify, Langflow):
ss -tlnpon each server plus an external scan of your IP ranges. Put them behind a VPN or an authenticating reverse proxy. - Replace every default secret, including the LiteLLM master key and admin accounts created at install; finish or remove any half-configured instance.
- Update Langflow and RAGFlow to the versions fixing the CVEs above.
- Search for secrets in
.envfiles, front-end bundles, source maps and repositories, and rotate any AI or cloud key that may have leaked. - Restrict access to cloud metadata from applications and set spending alerts on your AI provider accounts.
Our take
None of these techniques is new; what changes is the speed, as an agent finds and exploits the same mistakes in hours, around the clock. An AI tool deployed in a hurry by a business team is a server like any other and needs the same hardening, patching and monitoring.
Sources
- FBI FLASH-20261113-001 — Threat Overview of ARTEX Autonomous Penetration Testing Framework
- AWS Security Blog — Amazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents
- The Record — FBI flash alert says Artex tool automating cyberattacks at machine speed
How we choose, verify and write our analyses →
Is your own server exposed? Run a free check — public information only, no intrusion:
Worried your servers are exposed? Our team audits, patches and monitors Linux and cloud infrastructure 24/7.
Talk to an expert Our managed server services →

