Cloudflare: AI pentesting tools exploded, the defense model is shifting

What happened?

Cloudflare said the number of large language model (LLM) pentesting and orchestration frameworks has surged over the past month. According to the company, more than 50 framework families attempted to scan Cloudflare customers globally in just the last three days; the number of IP addresses running these frameworks has reached the thousands and keeps growing daily.

ARTEX and its offshoots

ARTEX, an open-source agentic pentesting system, was the only framework available a few weeks ago. Its creator, Autumn-27, announced on October 8 that the tool was being taken down. Since then, more than 12,000 repositories have appeared, with hundreds of forks carrying traces of the original code. Daily requests to AI pentesting and exploitation tools have grown 30x since October 1.

How the attacks work

These frameworks combine existing connections such as the Model Context Protocol (MCP) with traditional pentesting tools, letting AI agents carry out stages of the cyber kill chain at scale. ARTEX and its forks map environments onto a graph, allowing multiple independent agents to enumerate, verify and attack in parallel. Cloudflare notes that some complex attacks also chain stolen customer credentials.

The six recommended pillars

Cloudflare's 'Modern Security Model' includes:

• Fortify the front gate: Use a Web Application Firewall (WAF) with advanced rules such as rate limits.
• Security station in the foyer: Apply anomaly detection to all web applications customers access.
• Identify phony visitors: Implement bot defense.
• Remove the back door: Use identity-gated access for high-risk external and internal applications.
• Inspect deliveries: Enforce validated schemas and explicit security controls.
• Neighborhood watch: Use threat intelligence for visibility into attacks.

Cloudflare expects these frameworks to keep proliferating and says every organization should be prepared for the threat.