Fortinet introduces the FortiGuard Inline AI Security Service. This service enables organizations to monitor AI interactions through their existing FortiGate firewalls.
The Inline AI Security Service is offered as a cloud service and is designed to detect AI-specific threats. The service also aims to prevent the exposure of sensitive data and monitor autonomous AI agents, without requiring separate security infrastructure.
FortiGate firewalls have long been able to identify unauthorized AI use (“shadow AI”) and apply data protection policies to AI traffic. The new service goes a step further by inspecting the content of interactions between users, AI models, and agents.
The technology is based on FortiAIGate, Fortinet’s security platform for AI applications and large language models (LLMs). The service also utilizes technology from Virtue AI, a company Fortinet acquired two months ago.
Inspecting AI interactions
Traditional network security tools primarily examine IP addresses, signatures, and network traffic for suspicious activity. However, AI applications pose risks that traditional inspection methods may not detect.
Malicious instructions can be hidden in prompts, model responses, or communications between AI agents and external tools. These instructions can cause agents to disclose confidential information, execute unauthorized commands, or gain access to systems beyond their intended permissions.
Fortinet’s new service inspects prompts, model responses, API activity, and communications in real time using the Model Context Protocol (MCP). The inspection covers text, images, audio, video, and code.
The service can also recognize 35 categories of personal and confidential information. Depending on the security policy, sensitive data can be blocked before it is sent.
Monitoring unauthorized AI agents
FortiAIGate provides additional control over AI usage. Administrators can restrict access to specific models, set spending limits, and monitor interactions between AI agents, applications, and MCP servers. The platform is also designed to prevent “Denial-of-Wallet” attacks, in which malicious requests consume computing power to generate exorbitant costs.
During Black Hat USA in August, Fortinet Executive Anthony James warned that employees are increasingly installing unauthorized AI agents and MCP servers on their computers. These systems could transmit corporate information to external AI providers, including to train models.