Cloudflare warns AI pentesting tools are accelerating attacks

Cloudflare reports a sharp rise in AI-driven penetration testing orchestrators, with a proliferation of large language model (LLM) based frameworks that automate offensive security workflows. Since ARTEX, the open‑source agentic pentesting system, was taken down by its creator on 8 October, the ecosystem has exploded: there are now more than 12,000 repositories with hundreds of forks, thousands of IPs seen running these tools, and daily requests from such frameworks up by 30x since 1 October.

Cloudforce One’s incident response investigations describe attacks where credentials stolen from CI/CD pipelines are chained with LLM exploitation tools to move laterally within victim environments, sometimes pivoting from hyperscalers into Cloudflare customer environments.

One ARTEX‑involved incident showed a compromised account‑level API token used for CI/CD with no IP restrictions, resulting in rapid activity across Cloudflare infrastructure, including 69 requests per minute and 400 database dump requests in an hour.

In response, Cloudflare endorses a Modern Security Model built around six pillars: Fortify the front gate with a WAF and rate limits; security station in the foyer via anomaly detection on web apps; identify phony visitors with Bot Defense; remove the back door through Identity‑Gated Access and micro‑segmentation; inspect deliveries by enforcing validated schemas and explicit controls; and maintain a neighbourhood watch with threat intelligence.

The company notes that ARTEX and forks now account for over 60% of known AI pentesting orchestrators, with ongoing growth in forked variants post‑October. The article emphasises containment and identity‑gated approaches as essential when LLM‑driven attacks can orchestrate multiple kill‑chain steps at speed.