Skip to Content

Inside ARTEX: What Defenders Need to Know About Agentic Offensive Frameworks

Security leaders have warned that AI agents would speed up attacks. That warning is now reality.

Attackers are now using agentic offensive frameworks (AOF) to target companies, financial institutions, and government agencies worldwide. These tools copy legitimate testing methods, using large language models to scan targets on their own, gain access to poorly secured networks, and carry out attacks on behalf of human operators. ARTEX is the clearest example.

What Is ARTEX?

ARTEX is an AI-powered penetration testing system designed to execute end-to-end security assessments. Attackers can also weaponize it as an AOF to gain access to an organization’s systems. ARTEX uses AI agents to continuously scan for vulnerabilities, test multiple attack paths, and switch tactics whenever an attempt fails.

What makes ARTEX significant is how quickly it can connect security gaps, such as exposed passwords, API keys, or misconfigured applications, to reach sensitive data or systems. While a human operator still steers the tool, AI automates the heavy lifting. This gives threat actors the speed and scale to target more organizations at once, allowing them to focus on building new exploits or negotiating with the organizations they are extorting.

How Attackers Are Using ARTEX

ARTEX only appeared openly on GitHub in July. In just a few months, it started showing up in real intrusions. Its developer pulled the repository on October 8 following abuse reports, but forks, archived copies, and a commercial edition called ARTEX PRO remain active. Researchers have identified more than 1,600 exposed consoles online. ReliaQuest has observed multiple threat actors using ARTEX since at least August, sometimes achieving privileged access in as little as two hours.

Most activity centers on external web attacks: SQL injection, file-upload flaws, and path traversal. Attackers start with heavy recon and OSINT, then find subdomains and directories. ARTEX goes straight for exposed API documentation and probes every endpoint it finds. It favors exposed or quickly built, “vibe-coded” applications.

In one case, an unauthenticated API with admin-level identity functions sat on the internet with no logging. In another, an endpoint issued tokens and exposed user data, including financial details.

What Makes ARTEX Harder to Catch?

Most users are opportunistic data-extortion groups, not ransomware groups. They steal data and encrypt nothing. ARTEX is a force multiplier: one attacker can hit every domain they can find, in bulk. Attackers pair it with guardrail-free models, often locally hosted, open-weight Chinese models or commercial uncensored services, to reduce AI refusals.

Detection is becoming more difficult as ARTEX evolves. Forks strip identifying traits, including the framework’s own user agents, which now look like ordinary tools. The base version has the most features and the most detectable traits. Older forks have fewer features but better operational security. Indicator-based hunting is decaying, so detection has to be behavioral.

What This Means for Defenders

ARTEX does not need a new kind of exploit. It finds and uses the gaps most environments already have. Defenders should prioritize three actions:

  • Inventory and harden internet-facing applications. Include test and staging systems that reach production, use Zero Trust access or a VPN where possible, and enable WAF blocking, rate limiting, and bot controls.

  • Remove exposed keys and rotate credentials. ARTEX hunts for secrets. Restrict admin pages, backups, source maps, and .env files. Apply least privilege and test registration and password-reset flows for ownership checks.

  • Cover every internet-facing server. Make sure EDR is active on each one, and that WAF, web server, identity, and cloud logs are forwarded to your security tools. Application records often show the first signs of an attack.

Watch for four signs: repeated automated requests, unexpected accounts or access changes, unfamiliar web-shell files, and unauthorized outbound data transfers.

The common thread is speed. To defend against AI attacks, defenders must find and fix gaps faster than an automated tool can find and use them.

GreyMatter Defends Against Attacks Like ARTEX

AOFs like ARTEX continue to show what attackers are capable of with AI. The ReliaQuest Threat Research team tests ARTEX and other AOFs to build behavior-based detections around the tools and TTPs attackers use.

GreyMatter Attack gives defenders the same advantage by allowing your team to run ARTEX-style attacks against your own environment before an attacker does and goes further by helping you remediate findings.

Pre-built campaigns replay the TTPs our researchers see in ARTEX and other AOFs, and new research becomes a new campaign. For a custom run, describe the scenario in natural language and Attack builds the campaign. Attack returns a visual attack path that shows every gap with prioritized recommendations and automatically closes related alerts to reduce noise caused by false positives.

Attack is native to GreyMatter, so findings flow back into the platform. Agentic Teammates use them to build detections and harden controls, keeping your team in control of what gets deployed.

Learn How GreyMatter Agentic AI Scales Your Security Operations

GreyMatter is an agentic AI security operations platform with 6 agentic Teammates that use hundreds of agent skills and AI tools to work toward an objective, not just tasks.

GreyMatter dashboard active summary