Adapting to AI penetration testing orchestrators: The Modern Security Model
The number of AI penetration testing frameworks has exploded in recent weeks, with over 50 families attempting to scan Cloudflare customers globally and thousands of IP addresses running these tools daily.
Just a few weeks ago, ARTEX was the only open-source agentic penetration testing system available. Now, the ecosystem has exploded with over 50 framework families attempting to scan Cloudflare customers globally. The number of IP addresses running these frameworks has grown into the thousands and continues to increase daily.
ARTEX was taken down by its creator on October 8, but this removal triggered a "hydra effect," with over 12,000 repositories now containing several hundred unique forks. Daily requests of AI penetration testing tools have grown by 30x since October 1, with these tools beginning to chain stolen customer credentials in complex attacks.
The AI Threat Landscape
These LLM penetration testing frameworks orchestrate AI agents to carry out stages of the traditional Cyber Kill Chain at scale. Tools like ARTEX can map environments onto graphs, allowing multiple independent parallel agents to enumerate, verify, prototype, and attack at the speed of inference. As of November 12, ARTEX and its forks make up over 60% of currently known AI penetration testing orchestrators.
The malicious exploitation of this tool runs entirely counter to its original intent.
Identity-Chained Exploitation
Cloudflare's incident response team has observed how exposed credentials are being chained with LLM-based exploitation frameworks. In multiple incidents, threat actors gained initial access to hyperscaler environments and pivoted directly into customers' Cloudflare environments through compromised static, non-expiring tokens in CI/CD pipelines.
In one case, ARTEX gained access through a compromised account-level API token with no IP restrictions or restrictions. The framework created numerous Worker scripts with names like "artex-*" and generated activity exceeding 69 requests per minute across multiple parts of the infrastructure, along with 400 database dump requests in an hour.
The Modern Security Model
To counter these AI-powered attacks, Cloudflare recommends organizations adopt the Modern Security Model with six key pillars: Fortify the front gate (implement WAF with advanced rules), Security station in the foyer (anomaly detection), Identify phony visitors (Bot Defense), Remove the back door (Identity-Gated Access), Inspect deliveries (validated schemas), and Neighborhood watch (Threat Intelligence).
Large Language Models increase a threat actor's ability to rapidly iterate through the Cyber Kill Chain against a target, giving them advantages of speed and validation that vaporize legacy cyber security models. Organizations must prepare for these autonomously guided attacks as the frameworks continue to proliferate.
Related
