ARTEXARTEX

Agentic AI penetration testing framework developed in China by Autumn-27: a multi-agent LLM system written in Go with a Next.js frontend, model-agnostic, that runs an autonomous test from reconnaissance to data exfiltration and accumulates discovered assets and vulnerabilities into a graph. Released as open source, it was made closed source by its developer after its abuse became public. CrowdStrike first reported its use, alongside Claude Code, DeepSeek, Grok and GLM, by an unattributed suspected Chinese-speaking, financially motivated operator against South Korean financial institutions from late September 2026, activity Recorded Future tracks as TAG-217. Microsoft reports that Storm-0940 used it from September 2026 to automate password spraying, reconnaissance and OAuth application abuse against hundreds of organisations, including downstream tenants in the ConnectWise cloud backup supply chain compromise, and that actors across multiple regions use it. Dual-use: its presence alone does not indicate malicious activity.