What Is Artex AI (2026)? FBI Warning and Defender Takeaways

By Antigravity Skills editorial team · Last updated

Short answer: Artex AI, written ARTEX in the FBI's advisory, is an open-source framework that uses large language models to run penetration tests with AI agents. On 13 November 2026 the FBI warned that criminals have adapted it to automate intrusions. The original repository was taken down in October 2026, but forks and a commercial edition remain.

Unofficial summary. Antigravity Skills is not affiliated with ARTEX or the FBI. Every fact below comes from one primary source, the FBI's FLASH alert FLASH-20261113-001 published through the Internet Crime Complaint Center (IC3), which we read on November 18, 2026. We did not read the original project's repository or documentation, because the developer removed the repository, so anything that only the project could confirm is marked as unverified. This page is about defence only: it contains no attack instructions and links to no copies of the tool.

What is Artex AI?

According to the FBI, ARTEX is an autonomous, multi-agent penetration testing platform written in Go. You describe a goal in natural language, and specialised AI agents turn it into machine actions with little further human input. Penetration testing normally means a security team probing its own systems with permission, and the project's developer said the tool was meant for authorized testing within a defined scope.

Two details matter for understanding what it is:

  • No model of its own. ARTEX does not ship an AI model. It connects to outside model providers. The FBI names Anthropic Claude and OpenAI GPT as well as DeepSeek, GLM, Grok and Kimi.
  • Agents do the slow parts. The FBI says the agents automate time-consuming stages such as enumeration and exploitation, so a small group can run several campaigns at once.

Who made it, and is it still available?

The FBI says the tool was first released on GitHub in July 2026 by a developer who goes by Autumn-27, and reached an estimated 1,500 to 2,000 stars within ten weeks. After reports of abuse, the developer made the repository unavailable on October 8, 2026. The alert still describes the framework as spreading through community forks, localized variations, archived copies and a commercial version called ARTEX PRO.

The FBI also says that as of early November 2026 researchers found more than 1,600 publicly exposed ARTEX console instances, and that the real number is probably higher because many sit behind firewalls.

What license is it under?

We could not verify this. The FBI alert calls ARTEX "open-source" but does not name a license, and the original repository is gone. If a license matters to you, read it in the copy you intend to use, from a source you trust, rather than relying on a summary.

What did the FBI say?

The alert is an FBI FLASH that sets out the tactics and indicators tied to ARTEX. Its core message, in our paraphrase: a variety of malicious actors have used ARTEX since at least August 2026 against organizations worldwide, across sectors including financial services, government, technology, energy, healthcare and telecommunications. In the FBI's words, it lets people with little technical skill "conduct complex, multi-target operations simultaneously."

The FBI cites some intrusions moving from first targeting to privileged access in under three hours, and describes two notable campaigns:

  • An attack on South Korean financial institutions that exposed personal information of about 67,000 people.
  • A supply chain compromise affecting ConnectWise Cloud Backup customers, which led to attempted sign-ins to more than 2,250 downstream tenants.

FBI points compared with what we could check

We built this table from the FLASH text so you can see which claims come from where. The right column says whether we could confirm it beyond the alert itself.

FBI pointOur readingIndependently verified by us?
ARTEX is open-source, written in GoDescribed as an agentic penetration-testing platformNo, the repository is removed
Developer removed the repo on October 8, 2026Spread continues through forks and ARTEX PRONo, FBI statement only
Used since at least August 2026Many actors, many sectorsNo
Some intrusions reached privileged access in under three hoursSpeed is the main defender concernNo, FBI cites its own sources
1,600+ exposed consoles in early NovemberLikely an undercountNo
No license namedOpen-source status taken from FBI wordingNot stated by the FBI

The alert also includes lists of indicators of compromise. The FBI advises organizations to vet them before blocking anything, because cloud IP addresses get reassigned and an address once used for abuse may later serve harmless traffic.

What the FBI says it changes for defenders

The alert lists risks that follow from automation:

  1. Scale. Enumeration and exploitation that took people days can be run in parallel.
  2. Overlooked systems. Peripheral systems, such as partner and supplier portals, mobile backends, legacy services and applications from different business units, get less security attention and are easy targets.
  3. Supply chain reach. Weaker vendors and build pipelines can be a way into larger organizations.
  4. Lower skill bar. Less experienced actors can run complex operations.
  5. A shorter response window. The traffic is noisy, but the speed leaves little time to react.

Defensive takeaways

These are the FBI's recommendations, summarised and shortened. Read the full alert for the complete list.

Reduce what is exposed. List every internet-facing application, including partner portals, legacy services and third-party integrations. Restrict cloud administration, deployment and backup interfaces to trusted networks, and move enterprise applications behind zero-trust access where you can. Require an architecture review before a workload gets a public address.

Tighten identity. Apply least privilege to multi-tenant application permissions and consent grants. Watch for new credentials and privilege assignments on applications. If you used the affected ConnectWise integration, review every Microsoft 365 tenant where it was ever installed, searching by application ID, and revoke permissions and rotate credentials if you confirm suspicious activity.

Protect applications. Put rate limits and abuse controls on high-impact APIs. Alert on sequential record lookups on partner and employee portals, enforce server-side authorization, and avoid showing detailed error messages to unauthenticated users. Scan code, build artifacts and pipelines for embedded secrets, and rotate any that are exposed.

Preserve evidence and report. Keep sign-in, audit and SIEM records, and do not hold up urgent containment while waiting for exports. The FBI asks organizations to report incidents to a local field office or to IC3, including signs of unexpected outbound calls to model providers or unfamiliar model proxies in egress logs.

Why a coding-agent site is covering this

The same pattern, an AI agent allowed to run commands for you, is behind everyday tools. If you use agents yourself, the lesson from the alert is to control what they can reach. Our guide to Claude Code's "Don't ask again" permission explains how broad allow rules widen what an agent may run. OpenClaw is another open-source, self-hosted agent whose security notes are worth reading before you connect accounts. To use an agent for defence on your own code, see the Cloudflare Security Audit skill. For agent harness design in general, see DeepSeek Harness, a different open-source project that the FBI alert does not mention.

Sources

What Is Artex AI (2026)? FBI Warning and Defender Takeaways