Even approved AI tools can expose confidential information when employees use them.
Fortinet announced the FortiGuard Inline AI Security Service on Oct. 26, bringing cloud-delivered AI protection to existing FortiGate firewalls. Customers can activate the offering without installing separate AI security infrastructure.
For partners managing FortiGate environments, the announcement raises practical questions about which customer AI workflows can be inspected and who will investigate the resulting alerts.
New inspection examines prompts and model responses
FortiGate already offered detection of shadow AI applications and traditional data loss prevention (DLP) policies for AI traffic. Fortinet says the FortiGuard Inline AI Security Service adds AI-aware inspection to existing FortiGate deployments.
The service uses FortiAIGate technology, which Fortinet says detects prompt injection and jailbreak attempts designed to manipulate model behavior. FortiAIGate’s broader inspection capabilities cover model prompts and responses, Model Context Protocol (MCP) tool communications, API activity, and agent actions. Policies can account for the model, application, and user’s role.
A prompt containing customer account numbers, for example, may disclose protected data through an otherwise legitimate chatbot. Depending on configured policies, safeguards can alert administrators, mask sensitive details, or block the request before transmission.
FortiAIGate expands governance and deployment options
Aside from the firewall service, Fortinet also detailed enhancements to FortiAIGate, the existing gateway that supplies its AI inspection technology. Organizations that deploy the gateway separately gain additional governance capabilities, with guardrails strengthened by technology from the company’s Virtue AI acquisition.
Administrators can assign model permissions, spending limits and request rate limits by organization, workspace, project or user. Budget alerts warn when usage approaches a threshold, and requests are blocked once spending limits are exceeded. Fortinet says intelligent routing directs requests to models based on content, policy, and cost. Its denial-of-wallet controls are designed to block malicious requests before they consume tokens or GPU resources and drive up costs.
Dedicated workspaces allow business units to maintain separate policies under central administration. Visibility into AI agents reveals their connections to business applications and MCP servers. More than 50 customizable compliance policies and detailed logs of prompts, outputs, and tool calls support auditing and investigations.
Expanded safeguards also cover multimedia and code, with DLP detection for 35 types of personal and confidential information. Customers with stricter data-handling requirements can host the gateway on premises or in their own cloud using Kubernetes and NVIDIA GPUs, keeping prompts and logs within their environment.
AI security expansion creates managed service opportunities for partners
Existing Fortinet-based managed network security engagements give channel partners an opportunity to extend customer protection into AI applications. That said, activating inspection alone does not establish how much AI traffic is covered or how customers will manage the resulting security alerts.
Responsibilities will differ depending on the services partners already provide.
- MSPs managing FortiGate environments. Understanding which AI applications send traffic through existing inspection points is the first consideration. Providers can then test data protection policies against routine customer workflows, identifying coverage gaps and legitimate requests that might be blocked before wider deployment.
- MSSPs delivering security operations. AI-related alerts introduce additional investigation and response work. Providers can incorporate suspicious prompts and potential data disclosures into existing monitoring services, with customer-specific escalation procedures and reporting that distinguishes actual threats from false positives.
- VARs and systems integrators. Deployment requirements will influence whether customers use the cloud-delivered service or self-hosted FortiAIGate. Partners working with organizations that have stricter data residency requirements should evaluate the appropriate architecture, licensing and FortiGate compatibility before finalizing project scopes.
As customers connect more AI agents to internal business systems, existing policies will need reassessment as permissions and application access change. Recurring security reviews offer partners a way to maintain those protections as customer AI environments develop.
More news: Oracle is expanding its OpenAI partnership with new tools that connect ChatGPT and Codex to enterprise applications, databases, and cloud infrastructure.