by Chris DePuy, 650 Group / October 26, 2026
This morning, Fortinet announced the FortiGuard Inline AI Security Service, a cloud-delivered service that brings AI-specific inspection and guardrails to FortiGate firewalls without new infrastructure. The service is built on the company’s FortiAIGate technology and draws on capabilities from the recent Virtue AI acquisition, per the announcement on Fortinet’s investor site. We think the distribution move is the real story: Fortinet is folding AI-gateway functions into a firewall base it already has installed, while the depth of the inspection underneath remains largely undisclosed.
What did Fortinet announce for AI security on FortiGate?
The announcement covers two deployment paths. Customers running FortiGate can enable the FortiGuard Inline AI Security Service as a cloud-delivered subscription with no extra AI security stack to stand up or run. Organizations with stricter privacy or sovereignty requirements can instead self-host FortiAIGate on Kubernetes with NVIDIA GPUs, keeping prompts, responses, data, and logs in-house. Buyers standardized on Fortinet get the fastest path — turn on a service on firewalls they run today — while enterprises that cannot let AI traffic leave their environment get the self-hosted gateway. Both routes draw on the same FortiAIGate foundation, per Fortinet’s release.
What the service inspects
FortiAIGate sits as a real-time control point between AI applications and models. The guardrails watch prompts and model responses, traffic on the Model Context Protocol between agents and tools, API calls, and the actions agents take, and Fortinet says the engine handles five content types: text, image, audio, video, and code. The service extends what Fortinet already offered for discovering shadow AI usage and applying DLP to AI traffic: rather than only flagging AI use and sensitive-data movement, it now reads what is said inside AI sessions and can mask sensitive fields before they leave. The expanded DLP engine detects 35 types of personally identifiable information and confidential data, and administrators can alert, deny, or redact before exposure, per the release.
How spend controls and agent visibility work
The governance layer reads like the spec sheet of a standalone AI gateway. Administrators set model permissions, budgets, spend limits, and rate limits at the organization, workspace, project, and user level; the platform alerts as spending approaches a limit and blocks requests once it is exceeded, and Fortinet says malicious prompts and denial-of-wallet attacks are stopped before they consume tokens or GPU compute. Intelligent routing directs each request to a model based on content, policy, and cost without application re-architecture. Dedicated workspaces give teams their own policies, models, logs, roles, and permissions under centralized control, and more than 50 predefined compliance policies can be added with one click, with detailed logs of every prompt, output, and tool call for auditors. A visibility layer shows which AI agents operate across the business and how they touch enterprise applications, APIs, tools, data, and MCP servers, per Fortinet’s announcement.
Cloud delivery or self-hosted gateway
Fortinet frames the choice as one of cost, performance, privacy, and sovereignty. Founder, President, and CTO Michael Xie said in the release that Fortinet is “making AI security available through the infrastructure customers already use,” with FortiAIGate as the foundation for the broader strategy. The self-hosted path matters for enterprises where prompt and response data cannot cross the enterprise boundary; the release states prompts, responses, data, and logs stay inside the environment when FortiAIGate runs on the customer’s own Kubernetes cluster with NVIDIA GPUs. The company ties the service into its Security Fabric, connecting AI controls to the network, cloud, application, data, identity, and security-operations sides of the platform, per the announcement.
What Fortinet did not disclose
While the company did not disclose accuracy or false-positive rates, it claims minimal false positives and low latency. The company didn’t disclose how the inspection models themselves were built, trained, or updated, but given that the cybersecurity market is a game of cat and mouse, perhaps it is better that Fortinet keeps these details to itself. What is very clear is that the pace of AI attacks is growing and advanced tools like Fortinet’s AI Inspection are more important than ever before. Inline inspection of prompts, agents, and tool calls is becoming a standard line item in the security platform, and Fortinet’s move puts that function where the enterprise network already sits.
650 Group researches several of the technologies we covered in this blog, including Security and SASE, Enterprise, and more.
