AI pentest traffic against Cloudflare customers is up 30x
Cloudflare says daily requests from AI-driven penetration-testing tools have grown 30-fold since 1 October, and the forks keep multiplying.
Cloudflare says daily traffic from AI-driven penetration-testing tools hitting its customers has grown 30 times over since 1 October. The post, published 13 November, traces the surge back to ARTEX, an autonomous, multi-agent pentesting and command-and-control framework that its creator pulled down on 8 October.
Taking the original repository down backfired: there are now over 12,000 related repositories and more than 50 distinct framework families actively scanning Cloudflare’s customers, with ARTEX and its forks still making up over 60% of known AI pentest orchestrators. Cloudflare logged one attack hitting 69 requests a minute and 400 database dump attempts in an hour, some chaining in stolen credentials for lateral movement.
Why it matters: these tools enumerate, verify and attack in parallel at inference speed, which is a different load pattern to the manual scanning most rate limits and WAF rules were tuned against. Cloudflare’s advice is the unglamorous basics done properly: WAF rules with real rate limits, bot detection switched on, identity-gated access instead of open endpoints, and API schema validation. Worth checking those are actually configured, not just available on your plan.
The caveat: Cloudflare doesn’t name victims or give a total breach count, only traffic volume and repository counts, so there’s no way to tell from this post how many of these scans are succeeding.