Cloudflare Sees Surge in AI Pentesting Scans, Recommends Modern Security Model

· Original article ↗

Summary

Cloudflare reports 50+ AI pentesting frameworks scanning customers and a 30-fold rise in requests since Oct. 1. It says exposed CI/CD tokens enabled cross-environment access and recommends WAF, bot defense, identity gating, microsegmentation and schema validation.

Key points

  • Cloudflare says more than 50 AI pentesting framework families attempted to scan its customers in three days, with the number of observed IP addresses in the thousands.
  • Daily requests from these tools have risen 30-fold since Oct. 1; Cloudflare says ARTEX and its forks accounted for over 60% of known orchestrators as of Nov. 12.
  • Cloudflare's Cloudforce One team says it supported multiple investigations involving ARTEX-derived frameworks and observed attackers pivoting from hyperscaler environments into customer Cloudflare environments.
  • Cloudflare attributed a recurring access problem to compromised static, non-expiring CI/CD tokens with broad access across environments.
  • In one investigation, Cloudflare observed ARTEX using a compromised account-level API token, followed by new Worker scripts, more than 69 requests per minute, 400 database dump requests in an hour, and attempted persistence.
  • Cloudflare recommends a six-part defense model: traffic controls and WAF, anomaly detection, bot defense, identity-gated access and microsegmentation, validated API schemas and explicit security controls, and threat intelligence.

Article Details

Defense Focus
Detect and contain automated AI-assisted reconnaissance and exploitation, while limiting the impact of compromised credentials through layered access controls and segmentation.
Detection Methods
  • Monitor for automated scanning patterns, unusually high request rates, abnormal request paths, and mechanical bot behavior.
  • Watch for unexpected Worker script creation, including names matching the observed artex-*, artex-pgrecon-*, artex-pgw-*, and artex-do-* patterns.
  • Investigate spikes in API or database activity, including unusually frequent database dump requests, session-token validation attempts, and persistence attempts.
  • Review use of long-lived CI/CD tokens for access across cloud environments and customer accounts.
Data Sources
  • Web and network request activity
  • Worker script creation and activity
  • API and database request activity
  • Session-token validation and CI/CD token use
Rule Types
  • WAF rate-limit and anomaly-detection rules
  • Bot behavior detection and challenge rules
  • OpenAPI schema-validation rules
  • Mutual TLS certificate-validation policies
  • Explicit default-deny rules
Platforms
  • Web Application Firewall (WAF)
  • Bot Defense and behavioral challenges
  • Identity-gated access and Zero Trust controls
  • Micro-segmentation
Defensive Actions
  • Route incoming traffic through distributed traffic controls and apply WAF rate limits and anomaly detection.
  • Use bot detection and challenges to identify and block automated reconnaissance.
  • Require identity-gated access for high-risk applications and apply micro-segmentation to restrict lateral movement.
  • Validate incoming requests against OpenAPI schemas, require valid mTLS certificates, and reject requests that fail validation.
  • Use hardware FIDO security keys and monitor or restrict long-lived, broadly scoped CI/CD tokens.
  • Apply threat intelligence to update detections and blocklists as new infrastructure and techniques emerge.

MITRE ATT&CK

People

Vendors

Tools

Related Articles