Fortinet (NASDAQ: FTNT) has introduced the FortiGuard Inline AI Security Service, a cloud-delivered offering that adds AI-specific inspection and guardrails to FortiGate environments. The service is designed to help organizations examine AI activity and enforce data-protection and usage policies through security infrastructure they already operate, rather than requiring a separate AI-security deployment.
The change addresses a practical problem for security teams. Large language models and autonomous agents increasingly reach business applications, data and workflow tools, but conventional network controls are not designed to assess the meaning of a prompt, a model response or a Model Context Protocol tool call. Fortinet said its new controls inspect those AI interactions in real time and apply policy according to the model, application, user role and business context.
For an organization already using FortiGate, the cloud-delivered service can be enabled without deploying and managing separate AI-security infrastructure, according to the company. Fortinet is positioning that option for teams that want broader visibility into AI use and a way to block AI-specific threats or sensitive-data exposure through an existing security control point. Customers with more stringent privacy or sovereignty requirements can instead deploy FortiAIGate in their own environment.
FortiAIGate supplies the underlying guardrails for the service. Fortinet said it can inspect prompts and data across interactions with agents, large language models and MCP tools, and can enforce security, compliance and data-protection policy. The company said the service extends beyond shadow-AI discovery and data-loss-prevention policies for AI traffic by inspecting the content of interactions and masking sensitive data before exposure.
The announced controls span prompts, model responses, MCP tool calls and responses, API activity and agent actions. Fortinet said its data-loss-prevention functions can detect 35 types of personally identifiable information and confidential data, then alert, deny or redact that information. These are vendor-stated capabilities, and an enterprise will still need to determine its own classifications, policy thresholds and acceptable false-positive tradeoffs before relying on them in production.
The platform also includes policy controls aimed at model use and cost management. Administrators can set model permissions, budgets, spend limits and rate limits by organization, workspace, project or user, the company said. Fortinet said the system can alert administrators as spending approaches a limit and block requests after a limit is exceeded. It also said malicious prompts and denial-of-wallet attacks can be stopped before consuming tokens or GPU capacity.
Governance functions are another focus. Dedicated workspaces are intended to let teams maintain separate policies, models, logs, roles and permissions while a central team retains visibility. Fortinet said the product includes more than 50 predefined policies across AI-governance frameworks, privacy laws, industry regulations and security frameworks, with logs for each prompt, output and tool call. The service can also show which agents are interacting with enterprise applications, APIs, tools, data and MCP servers.
Fortinet is tying the service to a broader argument that AI controls should sit close to the paths that carry AI traffic. The approach may simplify adoption for existing FortiGate users, but it does not make AI governance automatic: security teams still need to set approved-model rules, review agent permissions and decide when a policy violation should block work. The new service gives them an inline enforcement layer for that work instead of treating AI traffic as an opaque extension of standard web traffic.