FBI FLASH on ARTEX: AI attack framework, 3 hours to admin
The FBI says the ARTEX AI attack framework takes intruders from first scan to privileged access in under three hours. Its fix list starts with your forgotten apps.
The FBI has put the ARTEX AI attack framework in writing. FLASH-20261113-001, dated November 13, says some ARTEX intrusions went from initial targeting to privileged access in under three hours. The bureau's first recommendation is not a product. It is a list: every internet-facing app you own, including the ones you forgot.
What actually happened
ARTEX is an open-source, multi-agent penetration testing platform written in Go. You give it a goal in plain language. Its agents turn that into commands. It has no model of its own and plugs into outside LLM providers. The FBI says attacker-run instances mostly use DeepSeek Flash models, for the price and for the "lack of observable guardrails when framed for penetration testing."
The numbers in the FLASH:
- Since at least August 2026: malicious use across ten sectors, among them financial services, healthcare, manufacturing and legal.
- About 67,000 people: personal information taken in attacks on South Korean financial institutions.
- More than 2,250 downstream tenants: reached through the ConnectWise Cloud Backup supply chain compromise.
- Over 1,600 exposed ARTEX consoles counted by researchers in early November. The FBI says the real total is likely much higher.
The developer pulled the GitHub repository on October 8. It did not matter. The FBI says forks, archived copies and a paid "ARTEX PRO" version keep it in circulation.
Read the initial access list closely. There are no zero-days in it. The agents got in through SQL injection, missing authentication on web APIs, anonymous file upload, weak administrative credentials, and hardcoded administrator passwords in JavaScript bundles.
Why the ARTEX AI attack framework matters for your business
The target is the app nobody watches. The FBI says ARTEX "excels" at peripheral systems: partner and supplier portals, mobile backends, employee apps, legacy services. Your main store is hardened. The quote form a contractor built in 2021 is not. That is the door.
Old bugs, new speed. Nothing on that list is new. The change is that a small crew can now run the boring enumeration against hundreds of targets at once. Being too small to bother with is no longer a defense.
Speed makes rate limits a security control. The FBI tells you to put rate limits and transaction thresholds on high-impact business APIs, and to alert on sequential record lookups in portals. An agent that pulls customer records one ID at a time looks nothing like a human. Catch the pattern.
Stop handing out the map. The FLASH says to keep stack traces, SQL errors and internal paths away from unauthenticated users. Agents read error pages faster than you do.
Search for secrets in what you ship. Scan repos, pipelines and build output for embedded credentials. Rotate what you find. We covered the exposed-console side of ARTEX before; this is the same lesson from the victim side.
Key takeaways
- FBI FLASH-20261113-001: some ARTEX intrusions reached privileged access in under three hours
- Initial access used known bug classes and weak credentials, not zero-days
- Peripheral apps are the favored target: partner portals, mobile backends, legacy services
- FBI fixes: inventory internet-facing apps, rate-limit business APIs, hide error details, scan for embedded secrets
- Removing the code from GitHub did not stop it; forks and a paid version still circulate
Do you know every app you have on the internet? We inventory the forgotten portals, add rate limits and server-side authorization, and pull secrets out of the build. Ask us for an attack-surface review, or see what we build.
Sources: FBI FLASH-20261113-001 (IC3).
- #security
- #artex
- #ai-agents
- #fbi
- #attack-surface
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Qwen4-Max released: $1.50 input, cost per task down 62%
Alibaba released Qwen4-Max with a 1M-token context and API access at $1.50 input and $4.50 output. It is not open weights. Measure cost per task before you switch.
Read itAI accounting software: Blue J $176M, Sumary $25M
AI accounting software raised two large rounds: Blue J $176M for tax research and Sumary $25M for firm workflows. Ask your accountant which AI reads your books.
Read it