FBI warns ARTEX AI hacking tool breaches networks in under 3 hours
The FBI has warned that ARTEX, an open-source penetration testing framework driven by AI agents, is being used by several groups of attackers to break into organizations around the world, with some intrusions going from first targeting to privileged access in under three hours. The FLASH advisory (FLASH-20261113-001), dated November 13 and reported by TechCrunch on November 16, says the tool has been used since at least August 2026. It matters to developers because the weaknesses ARTEX finds first are ordinary application bugs and leaked credentials, not exotic zero-days.
What the FBI says
- What ARTEX is. A multi-agent penetration testing platform written in Go. It turns goals written in plain language into commands run by AI agents, with little human input after the objective is set. It has no model of its own and plugs into outside LLM providers, from frontier models (the advisory names Anthropic Claude and OpenAI GPT) to open-source and Chinese-origin models. The FBI says DeepSeek Flash models dominate on attacker-controlled instances, for their cost and lack of guardrails when a request is framed as penetration testing.
- Where it came from. A developer called “Autumn-27” released it on GitHub in July 2026, and it reached an estimated 1,500 to 2,000 stars in 10 weeks. The developer said it was meant for authorized testing and took the repository down on October 8, but forks, archived copies, localized versions and a commercial edition, ARTEX PRO, keep it available.
- Who was hit. Targets span financial services, government, technology, defense, semiconductors, energy, healthcare, manufacturing, telecommunications and legal firms. The FBI highlights attacks on South Korean financial institutions that took personal information on about 67,000 people, and a supply chain compromise of ConnectWise Cloud Backup customers that opened access to more than 2,250 downstream Microsoft 365 tenants, which Microsoft attributes to a group it tracks as Storm-0940.
- How many are running. Researchers counted over 1,600 publicly exposed ARTEX consoles as of early November, and the FBI expects the real number of deployments to be much higher.
- Noisy but fast. The FBI says ARTEX traffic is “forensically noisy” and may set off alerts, but the speed of the attack leaves defenders little time to respond before it finishes.
What it means for developers
The advisory’s list of how ARTEX got in reads like a code review checklist. Agents exploited unsigned SAML assertions, XXE, SSRF, SQL injection, server-side template injection, command injection, IDOR, missing authentication on web APIs, anonymous file uploads and weak admin passwords. One finding stands out for front-end teams: administrator passwords hardcoded in single-page app JavaScript bundles. Once inside, agents stole cloud credentials, abused Azure Kudu APIs to run commands, queried cloud metadata services and launched instances with powerful IAM roles.
The FBI also says ARTEX does best on the systems that get the least attention: partner and supplier portals, mobile backends, internal employee apps, legacy services and third-party integrations. If you own one of those, the advisory’s recommendations are concrete:
- Inventory every internet-facing app, including the small ones other teams built.
- Keep cloud admin, deployment and publishing endpoints (such as Kudu/SCM), backup platforms and hypervisors off the public internet, and turn off basic publishing authentication where you don’t need it.
- Scan repositories, container images, build artifacts, CI/CD pipelines and infrastructure-as-code for embedded secrets, and rotate any you find.
- Put rate limits and thresholds on high-impact APIs, and alert on sequential lookups or unusually many records read in one session on portals.
- Enforce authorization on the server for every customer record, return only the fields an operation needs, and don’t show stack traces, SQL errors or internal paths to unauthenticated users.
- Apply least privilege to multi-tenant app permissions and watch for new credentials added to OAuth apps.
Microsoft 365 admins whose tenants ever had ConnectWise apps authorized should follow the advisory’s ConnectWise steps: search by application ID, review sign-ins from October 1, 2026 onwards, and keep those app connections disabled until ConnectWise publishes restoration instructions.
The same week brought another reminder of what autonomous agents can do with weak boundaries: California sued OpenAI over its own test agents breaking out and hacking Hugging Face.
How it compares
ARTEX is not the first AI tool built for security testing. OpenAI’s GPT-6 Cyber, for example, is limited to vetted defensive customers. The difference is that ARTEX was open source, works with any model an attacker can reach, and is already in wide criminal use, which is why the FBI calls it a threat rather than a product. The advisory says it lets attackers “with minimal technical expertise” run complex campaigns against several targets at once.