Fortinet’s FortiAIGate is bringing AI security controls to enterprise workloads on Oracle Cloud Infrastructure (OCI), helping businesses inspect prompts, block sensitive data leaks, and control agent actions.

An Oracle announcement published October 27 outlines how customers can combine the self-hosted gateway with OCI Enterprise AI to protect applications connected to company data and tools.

The approach addresses a growing security gap as AI agents move beyond answering questions. Agents can query databases, access code, and call business tools.

Monitoring chat messages alone leaves other interactions unchecked, including tool responses that could carry hidden instructions or expose private information.

FortiAIGate runs as a containerized, GPU-powered reverse proxy between applications and AI models. It routes requests through security checks before they reach their destination. The gateway also inspects model responses, giving security teams a central place to apply rules across AI services.

Fortinet Secures Oracle Cloud AI Agents

Its Model Context Protocol (MCP) protection checks tool lists, calls, and responses in both directions. Fortinet says these checks detect prompt injection, data leaks, and harmful content in traffic between agents and MCP servers.

FortiAIGate and OCI reference architecture (source : Fortinet )

This extends inspection beyond the user’s original request to the information agents receive while completing tasks. That threat has surfaced in research covered by Cybersecuritynews.com on prompt injection through malicious MCP servers.

Researchers described how hidden instructions could redirect conversations or trigger tool use without clear user consent. Separate reporting on attacks targeting AI infrastructure detailed attempts to exploit exposed gateways and agent tools.

According to Oracle, FortiAIGate uses Virtue AI guardrails to inspect prompts and responses across nine languages. Built-in data loss prevention detects 35 types of sensitive information and can block or redact matches. Administrators can also describe custom detection rules in plain language.

Role-based workspaces separate team policies, model access, and logs while keeping central oversight. More than 50 preset policies cover privacy, security, and AI governance requirements.

Workspace policy changes require approval, while records of routed prompts, outputs, and tool calls support audits. These features help enforce rules but do not establish compliance.

The gateway also limits spending by organization, workspace, project, user, and virtual key. Requests stop when budgets run out. Rate limits restrict requests and tokens per minute.

At the same time, guardrails target denial-of-wallet attacks designed to consume paid AI resources. Content-based routing can send routine work to cheaper models.

The reference design places FortiAIGate on GPU-enabled Oracle Kubernetes Engine worker nodes. A hub-and-spoke setup can centralize inspection while applications operate across separate cloud networks.

According to Fortinet, OCI Enterprise AI provides managed model access, agent tools, and governance. Self-hosting keeps gateway logs and inspection data within the customer’s environment, while model requests still reach the selected provider.

Businesses must therefore check model routing and processing locations against their data residency needs. Protection also depends on sending relevant traffic through the gateway, rather than leaving alternate application paths outside inspection.

Give your SOC the alert context to stops incidents quickly. Integrate Threat Intel from 16K SOCs.

The post Fortinet Brings AI Agent Security to Oracle Cloud to Block Prompt Injection and Data Leaks appeared first on Cyber Security News.