Detect Deepfakesby Resemble AI
Agentic AI attack

Cloudflare agentic AI attack — Nov 2026

Cloudflare incident responders identified the use of the ARTEX AI agent framework to exploit a compromised API token for unauthorized database activity

Reported date
Nov 16, 2026
Target
Cloudflare
Agent type
Other AI agent
Agent role
Used by the attacker

The exact incident date was not established. This entry is dated by its source report.

Updated Nov 20, 2026 · 1 min read

What happened

Cloudflare incident responders identified a security incident involving the ARTEX AI agent framework, which was used to leverage a compromised account-level API token. The token, originally utilized for CI/CD processes, lacked IP restrictions or rate limiting, providing the attackers with a significant entry point into the environment. The compromise of this token is believed to have originated through a third-party cloud provider, though the specific method of theft remains undisclosed.

Once access was established, the ARTEX framework executed a series of unauthorized actions. Incident responders observed the creation of new Worker scripts within the account, identified by prefixes such as artex-, artex-pgrecon-, artex-pgw-, and artex-do-. These scripts corresponded to specific tasks assigned by the agentic framework. The activity included approximately 400 database dump requests performed within a single hour, occurring at a rate of more than 69 requests per minute.

Beyond the database queries, the agents attempted to validate session tokens for various third-party applications and sought to establish persistence within the Cloudflare environment. Cloudflare has not disclosed whether the database dump requests successfully returned any data, nor has it confirmed if any attempts to establish persistence were successful. The company noted that much of the activity appeared to be lightly guided by a human operator. This incident was part of a broader trend observed by Cloudflare, where multiple investigations involved the use of ARTEX and similar tools, often stemming from the use of static, non-expiring CI/CD tokens that granted broad access across customer cloud setups.

Evidence in the reporting

Incident evidence
Cloudflare's incident responders saw ARTEX gain access to a compromised API token
Agent involvement
a planner agent splits the goal into tasks for worker agents

Sources