Fortinet is introducing the FortiGuard Inline AI Security Service, a cloud service that adds AI guardrails and AI traffic inspection to firewalls in the FortiGate range. At the same time, the company is expanding its AI gateway FortiAIGate with features for cost management, compliance and visibility into AI agents. Some of these features are based on technology from Virtue AI, which Fortinet has acquired.
According to Fortinet, FortiGate users can activate the new service on their existing firewalls without having to implement and manage a separate AI security solution. The service uses existing Fortinet capabilities to detect shadow AI and apply data loss prevention (DLP) policies to AI traffic. On top of that, the company says the service inspects the content of AI interactions and masks sensitive data.
Michael Xie, founder, chairman of the board and chief technology officer of Fortinet, says businesses need AI security that keeps pace with the speed at which AI is being deployed within organisations. Xie describes FortiAIGate as the foundation of this strategy and says customers can tailor the gateway to their requirements for cost, performance, privacy and data sovereignty.
Inspection of prompts and model responses
Fortinet says traditional security tools inspect IP addresses, signatures and payloads, but do not analyse the prompts and responses of AI applications. FortiAIGate acts as a control point between AI applications and models. According to the company, the gateway inspects prompts and data in real time and applies policies to interactions with agents, large language models (LLMs) and Model Context Protocol (MCP) tools. These policies depend on the model, the application, the user’s role and the business context.
The inspection covers text, images, audio, video and code. According to Fortinet, the DLP feature recognises 35 types of personally identifiable information and confidential data, which can be flagged, redacted or blocked. The company also claims low latency and a limited number of false positives.
Budgets, governance and agents
Administrators can set model permissions, budgets, spending limits and rate limits per department, workspace, project or user. When usage approaches a threshold, an alert is triggered. Once the spending threshold is exceeded, requests are rejected. Fortinet says malicious prompts and ‘denial of wallet’ attacks are blocked before they consume tokens or GPU capacity. A routing feature sends each request to a model based on content, policies and cost, without the need to change the architecture of applications.
Teams and business units get their own workspaces with their own policies, models, log files, roles and permissions, which can be managed centrally. For compliance, FortiAIGate includes more than 50 predefined policies based on AI governance frameworks, privacy legislation, industry guidelines and security frameworks. Each policy can be customised. Auditors are given access to logs of every prompt, output and tool call. In addition, the gateway shows which AI agents are active and how they communicate with business applications, APIs, tools, data and MCP servers.
Self-hosting for data sovereignty
Organisations can host FortiAIGate on premises or in the cloud on Kubernetes, using NVIDIA GPUs. Prompts, responses, data and log files then remain within their own environment. Fortinet is aiming this option at organisations with stringent requirements for security, privacy and digital sovereignty.
Will Townsend, chief analyst at LoneStar Advisory & Research, sees an advantage in Fortinet bringing AI security, governance, cost management and agent visibility together in a single gateway, rather than leaving customers to use separate tools. According to Townsend, integrating Virtue AI’s technology delivers improvements in guardrails and performance.
This article was originally published in Dutch on MSP Business. Read the original →
