Fortinet extends FortiGate firewalls into AI security and adds new FortiAIGate features

Fortinet extends FortiGate firewalls into AI security and adds new FortiAIGate features

Network security company Fortinet Inc. today extended its FortiGate firewalls into artificial intelligence security with the launch of the FortiGuard Inline AI Security Service.

The service’s guardrails come from FortiAIGate. Fortinet also offers FortiAIGate on its own as a gateway that sits between applications and AI models, and the company detailed new features for it alongside the launch.

The new service is delivered from Fortinet’s cloud to firewalls that FortiGate customers already own, so they don’t have to deploy a separate AI security stack. Prompts headed to large language models, agents and Model Context Protocol tools are checked in real time.

Fortinet’s firewalls could already spot shadow AI use and apply data loss prevention rules to AI traffic. The service adds inspection of what those interactions actually contain, and sensitive data in them is masked as it passes through.

Organizations with stricter privacy or sovereignty requirements can still host FortiAIGate themselves on Kubernetes with Nvidia Corp. graphics processing units, either on-premises or in their own cloud. Prompts and logs in those deployments stay inside the customer’s environment.

Michael Xie, founder, president and chief technology officer at Fortinet, said AI security has to “scale with how quickly AI is being adopted across the enterprise.” He described FortiAIGate as the foundation of Fortinet’s broader AI security strategy.

The FortiAIGate changes are aimed at traffic that conventional security products were not built to read. Older tools look at IP addresses and signatures, Fortinet said, so a prompt hiding instructions for an agent gets through.

Each request through FortiAIGate is judged as it arrives, partly on the role of the person who sent it. Images, video and code are inspected along with text, and the expanded data loss prevention engine now catches 35 kinds of personal and confidential information before it leaves.

Administrators can cap what a team or even a single user spends, and the gateway stops passing requests along once that cap is hit. Fortinet said prompts built to run up compute bills are blocked before they use any tokens and that requests can be routed to a model based on content, policy and cost.

Turning on one of the gateway’s more than 50 built-in compliance policies takes a single click. Every prompt, output and tool call is logged for auditors. Agents operating across the business show up in their own view, mapped against what each one connects to.

Fortinet bought AI security startup Virtue AI Inc. in August, and technology from that deal is already part of the gateway’s guardrails. Will Townsend, chief analyst at LoneStar Advisory & Research, said the quick integration has improved both performance and the guardrails. He called FortiAIGate’s approach “elegant,” noting that customers without a combined gateway otherwise end up running different tools for AI security, governance and cost.

Anthony James, executive vice president of marketing at Fortinet, laid out the governance problem in an interview with theCUBE, SiliconANGLE Media’s livestreaming studio, at Black Hat USA in August. “Now what we’re seeing is customers want to know what people are doing with their AI infrastructure internally,” James said. The worry is employees standing up unsanctioned agents and Model Context Protocol servers on their own desktops, he said, which “could be sending data outside to train other models.”

Photo: Robert Hof/SiliconANGLE