Illustration of a blue browser window with an open magenta padlock and a purple stopwatch with speed lines

FBI Warns of ARTEX AI Attacks: What to Check on Your Site

A
Admin
·

On 13 November 2026, the FBI issued an alert about ARTEX, an AI-agent penetration testing tool that several attacker groups have used against organizations since at least August 2026. According to the FBI, some intrusions reached privileged access in under 3 hours. If you run a website or customer portal, the entry points it lists are worth checking.

What we know

  • The alert: an FBI "FLASH" (FLASH-20261113-001) dated 13 November 2026 and marked TLP:CLEAR, meaning it can be shared freely. TechCrunch reported on it on 16 November.
  • What ARTEX is: per the FBI, an open-source platform written in Go that turns natural-language goals into commands run by multiple AI agents with little human input. It has no model of its own: it plugs into external models, including Anthropic's Claude, OpenAI's GPT, DeepSeek, GLM, Grok and Kimi. The FBI says attacker-run instances mostly use Chinese models, led by DeepSeek.
  • Timeline: its developer released it on GitHub in July 2026 for authorized testing; the FBI dates malicious use to at least August 2026. The repository was pulled on 8 October 2026, but forks, archived copies and a commercial version called ARTEX PRO remain, according to the FBI.
  • Who was targeted: financial services, government, technology, defense, semiconductor, energy, healthcare, manufacturing, telecommunications and legal organizations, per the alert.
  • Cases cited: a campaign against South Korea's financial sector that took personal data on about 67,000 people, and a ConnectWise Cloud Backup compromise that gave access to more than 2,250 downstream tenants.
  • How widespread: in early November 2026, researchers found more than 1,600 ARTEX consoles exposed online; the FBI believes total deployments are higher.
  • Web entry points it lists: administrator passwords hardcoded in JavaScript bundles sent to the browser, APIs without authentication, SQL injection, anonymous file upload, weak admin credentials, and PHP and JSP web shells, files an attacker leaves on a server to get back in.
  • What it recommends: 20 mitigations. For websites, the key ones are inventorying everything internet-facing, keeping admin interfaces off the public internet, not showing technical errors to unauthenticated visitors, rate-limiting portals, and scanning configs, scripts and repositories for leftover secrets.

What changes and what doesn't

  • No new bug to patch. The alert does not announce a vulnerability: it describes a tool that tries well-known mistakes very quickly.
  • Response time shrinks. The FBI describes a "compressed detection window": ARTEX traffic is noisy and may trip alarms, but attacks move faster than many teams respond.
  • The bar drops. Per the FBI, low-skill actors can run sophisticated intrusions, and small groups can hit many targets at once.
  • What the alert doesn't say: it gives no total victim count and doesn't single out small websites. Its cases involve large organizations. But the entry points it describes are the same ones a poorly locked-down business site has.
  • The indicators are historical. The FBI asks defenders to treat its IP addresses as past indicators: blocking them is no substitute for closing the doors.

How to check if it affects you

  1. List everything that is online. Your main site, old test or campaign subdomains, hosting control panel, phpMyAdmin, customer portal and any API. Forgotten systems are what an agent finds first.
  2. Take admin panels off the public internet. WordPress or Drupal admin, the hosting panel and phpMyAdmin should sit behind an IP allowlist or VPN where possible, with two-factor authentication.
  3. Search the code your visitors receive for passwords. Use "View source" and open your site's .js files: look for "password", "admin", "token" or "key". No credential should ever reach the browser.
  4. Don't let your site show its errors. A visible PHP or database error tells an attacker what is behind it. In WordPress, keep WP_DEBUG_DISPLAY off in production; in Drupal, set "Error messages to display" to "None".
  5. Check upload folders. wp-content/uploads or sites/default/files should not contain .php files. If you find one you don't recognize, treat it as an incident: keep a copy before removing anything and get help.
  6. Rate-limit attempts. The FBI cites high-frequency, sequential automated requests against login and search endpoints. Limiting attempts on your login and portal search slows an agent trying thousands of combinations.
  7. Rotate anything that was exposed. Check .env files, backups sitting in the public folder and GitHub repositories. Treat anything that was exposed as stolen: issue new keys and revoke the old ones.
  8. Read your logs. If your server doesn't use AI but makes outbound calls to model APIs (the FBI cites api.anthropic.com and api.deepseek.com), or you see bursts of requests to paths that don't exist, investigate. Preserve logs before changing anything.

Related: AI Agents Hunt Exposed AI Apps to Steal Keys: What to Check

Sources

Updates: if anything changes and a primary source confirms it, it will be added here.