Security1 hOctober 7, 2026, 14:21

Cloudflare Uses AI Harness to Probe and Harden Its WAF

Cloudflare placed frontier AI models in a controlled harness to test its WAF: across 45 scenarios the system generated 1,107 attack attempts, leaving 49 findings after human triage. The work led to three changes in Cloudflare's Managed Ruleset, including two new SSRF detections.

InfoQ

Open source

45 scenarios, 1,107 attempts, 49 findings after human triage
48 of 49 findings involved command injection or SSRF
Added SSRF - Obfuscated Host and SSRF - Restricted Protocol rules
Models ran black-box, without access to WAF rules or source code

#Cloudflare

South Korea probes whether AI agents were used in recent bank hacks

Security

AvisLoader: New Windows Loader Uses Tox P2P Network for C2

Google Cloud's RRSI uses regularization to stop agent harnesses overfitting benchmarks

Policy

Cloudflare Uses AI Harness to Probe and Harden Its WAF

South Korea probes whether AI agents were used in recent bank hacks

AvisLoader: New Windows Loader Uses Tox P2P Network for C2

Google Cloud's RRSI uses regularization to stop agent harnesses overfitting benchmarks

Paris Prosecutors Open Criminal Probe into Smart Glasses Used to Film Women