2 min Security

Cloudflare tries to reduce attacks on LLMs with new firewall

Cloudflare tries to reduce attacks on LLMs with new firewall

Cloudflare is developing “Firewall for AI”. The solution specifically addresses security issues that arise for companies getting started with AI. In addition, new security solutions are coming to address risks specific to a company.

LLMs and AI solutions can be found in more and more products and thus are entering more and more companies. Technologies don’t come without risks, and this is no different for AI. Cloudflare is capitalizing on these developments with its “Firewall for AI” solution. The solution gives security teams the tools to quickly discover new threats, Cloudflare said.

Part of Workers AI

‘Firewall for AI’ is a free component of Cloudflare’s Workers AI offering. That offering targets developers who want to deploy AI models. Through Workers AI, the underlying LLM runs at a location as close to the enterprise as possible.

Those developers can then choose to add “Firewall for AI” to the AI model, which automatically scans prompts from users for malpractice. These could include requests from the user who wants to extract specific data from the model. The solution additionally protects against prompt injection. This type of attack is specifically aimed at bypassing set limits in the model to prevent a chatbot from using swear words, for example.

Customized security

Companies that are still somewhat uncertain about AI can still engage with the technology. After all, cybercriminals are increasingly exploiting the possibilities of artificial intelligence for their activities.

In addition to hackers, however, security solutions also benefit from the technology. Cloudflare is going to do just that, using AI for improved security. The improvement should come as AI will map traffic patterns for each company to better tailor security solutions to the needs of the specific business. In other words, the defence strategy will be customized for the company.

“Fighting AI with AI is now a non-negotiable,” said Matthew Prince, Cloudflare’s chairman of the board. “a personalized approach to protect data and defend against complex threats unique to an organization’s attack surface, at-speed, and scale, is paramount.”

Also read: State hackers actively misuse AI for debugging and malware creation