Cloudflare unveils Firewall for AI to safeguard Large Language Models from abuse

Cloudflare has revealed its latest project, Firewall for AI, a security layer designed to protect Large Language Models (LLMs) from potential abuses. This advanced Web Application Firewall (WAF) is specifically engineered for applications utilizing LLMs, and includes tools that can be deployed to detect vulnerabilities and provide visibility for model owners.

Included in Firewall for AI's toolkit are features already part of WAF's offering, such as rate limiting and sensitive data detection. Additionally, a new protection layer is currently under development. According to Cloudflare, this new validation will scrutinize the prompt submitted by the end user, identifying any attempts to manipulate the model to extract data or other forms of abuse.

The deployment of Firewall for AI will mirror that of a traditional WAF. Each API request containing an LLM prompt will be scanned for potential attack patterns and signatures. Importantly, this firewall can be utilized not only for models hosted on Cloudflare's own Workers AI platform, but also those hosted on any other third-party infrastructure.

You can sign up here to join the waiting list and get notified when the feature becomes available.