Cloudflare has officially transitioned its AI Security for Apps from beta to general availability, introducing a suite of tools designed to safeguard the increasingly complex world of LLM-powered services. Unlike traditional web applications that follow fixed rules, AI agents process unpredictable natural language, making them vulnerable to prompt injection—where attackers trick models into leaking data or performing unauthorized actions.

A standout feature is the now-free endpoint discovery tool, which automatically scans a company’s web properties to identify every instance of an AI model in use. This shadow AI detection ensures security teams are not flying blind as developers rapidly integrate new models into their apps. Cloudflare identifies these points not just by their names, but by analyzing traffic behavior patterns to find hidden LLM connections.

The update also brings custom topics detection, allowing businesses to set specific boundaries for their models. For example, a healthcare firm can flag any prompt touching on sensitive patient data, while a retailer can block mentions of competitor products. By integrating these signals directly into the existing Web Application Firewall (WAF), companies can block or log malicious attempts using the same interface they use for standard cyber defense.

To further broaden its reach, Cloudflare is collaborating with IBM Cloud and Wiz. These partnerships aim to provide a unified view of an organization's security posture, tracking everything from the cloud-hosted models themselves to the edge-based guardrails that protect the final user interaction.

Cloudflare has officially transitioned its AI Security for Apps from beta to general availability, introducing a suite of tools designed to safeguard the increasingly complex world of LLM-powered services. Unlike traditional web applications that follow fixed rules, AI agents process unpredictable natural language, making them vulnerable to prompt injection—where attackers trick models into leaking data or performing unauthorized actions.

A standout feature is the now-free endpoint discovery tool, which automatically scans a company’s web properties to identify every instance of an AI model in use. This shadow AI detection ensures security teams are not flying blind as developers rapidly integrate new models into their apps. Cloudflare identifies these points not just by their names, but by analyzing traffic behavior patterns to find hidden LLM connections.

The update also brings custom topics detection, allowing businesses to set specific boundaries for their models. For example, a healthcare firm can flag any prompt touching on sensitive patient data, while a retailer can block mentions of competitor products. By integrating these signals directly into the existing Web Application Firewall (WAF), companies can block or log malicious attempts using the same interface they use for standard cyber defense.

To further broaden its reach, Cloudflare is collaborating with IBM Cloud and Wiz. These partnerships aim to provide a unified view of an organization's security posture, tracking everything from the cloud-hosted models themselves to the edge-based guardrails that protect the final user interaction.