Fortinet and OCI combine AI security and governance to help organizations scale applications and agents securely

By Ali Bidabadi and Elena Sunshine

The enterprise AI landscape is rapidly moving from isolated experiments to production systems embedded in business workflows. As organizations give AI applications access to proprietary data, code repositories, and critical enterprise tools, security can no longer be an afterthought. Enterprise AI risks go beyond prompts and responses to include the actions agents take. Monitoring chat interfaces alone is no longer enough to protect enterprise tools, code, and databases.

Enterprises need granular, dynamic controls to monitor incoming prompts, outgoing tool calls, and automated actions for unsafe, malicious, or policy-violating behavior. To address these needs, customers can combine the power of OCI’s models, agent capabilities, and platform governance with Fortinet solutions that are available on Oracle Cloud Infrastructure marketplace. For example, FortiAIGate’s AI guardrails, cost controls, and runtime protection can help organizations scale AI while maintaining operational control. FortiAIGate introduces high-performance guardrails, cost control, role-based workspaces, and one-click compliance policies.

Balancing AI Autonomy with Enterprise Control

A recent Wall Street Journal report illustrates how increasingly capable AI agents can boost productivity while also exposing sensitive information, misinterpreting instructions, or taking unintended actions. In one case, an agent shared personal financial information in a company Slack channel, showing how quickly a routine task can become a security and privacy incident.

As development teams expand AI capabilities across applications, platform and security teams must establish consistent, centralized controls. Without a unified governance framework, teams may apply different permissions, safeguards, data-access policies, and monitoring practices to AI, creating gaps that attackers—or unintended AI behavior—could exploit.

Centralized controls establish clear guardrails for how AI systems access data, interact with applications, take action, and handle sensitive information. They also support consistent identity management, least-privilege access, auditability, and policy enforcement across the enterprise.

Embedding governance and security into the platform enables organizations to help developers innovate quickly while maintaining visibility, reducing risk, and protecting business-critical data. As AI capabilities expand faster, scaling security and governance alongside them becomes increasingly essential.

Building and Governing AI Applications with OCI

Customers can use OCI Enterprise AI with Fortinet solutions like FortiAIGate to help development teams prototype, scale, and launch AI agents while maintaining visibility, data governance, and security. OCI Enterprise AI consists of three integrated layers that help organizations build, orchestrate, and govern enterprise AI applications:

  • OCI Enterprise AI Models: Organizations can access and integrate multiple frontier models through a single interface, simplifying adoption and scaling of advanced models as needs and expertise grow.
  • OCI Enterprise AI Agents: Managed tools, standard protocols, and open-source frameworks help teams build, orchestrate, and deploy agentic workflows.
  • OCI Enterprise AI Governance: Security, compliance, access controls, and runtime guardrails support data protection and regulatory alignment across the AI lifecycle.

Moving an AI prototype to production introduces challenges related to latency, security, cost, and maintenance. OCI Enterprise AI addresses these challenges through four capabilities:

  • Managed model access: Oracle manages infrastructure provisioning and scaling. Dedicated AI clusters provide reserved model capacity to help deliver predictable performance and throughput.
  • Familiar APIs: OCI supports industry-standard API formats, including OpenAI-compatible interfaces, helping engineering teams adapt existing application logic with minimal code changes.
  • Open ecosystem: Support for open-source frameworks and open-weight models gives developers the flexibility to select and replace components as requirements change.
  • Ready-to-use runtime guardrails: Customers can adopt Oracle’s provided guardrails to help detect security risks and potentially harmful content.

FortiAIGate Secures and Governs OCI AI Apps and Agents

Alongside OCI Enterprise AI’s cloud-native controls, enterprises running production workloads need centralized security inspection and visibility across interactions among applications, AI agents, and model providers. FortiAIGate is a self-hosted AI gateway for large language models (LLMs), agents, and Model Context Protocol (MCP) tools. From a single control point, organizations can implement measures to block AI attacks and data leaks, enforce usage policies, and cap spending by user, team, and project.

Part of the Fortinet AI-Native Security Fabric, FortiAIGate runs on the organization’s infrastructure as a containerized, GPU-accelerated reverse proxy. Positioned between applications and models, it inspects prompts and responses and applies AI-specific guardrails to traffic passing through the gateway.

Two capabilities support this approach:

  • High-performance inspection: GPU acceleration and high-accuracy, low-latency guardrails enable inspection at production traffic volumes across predefined and custom policies.
  • Self-hosted deployments: Organizations can deploy the gateway in their own cloud infrastructure or private data center and retain control over inspection and gateway logs. Model requests are still sent to the providers they select.

FortiAIGate provides the following controls for AI applications and workflows:

  • Workspaces with roles: FortiAIGate uses organizations, workspaces, and projects to organize users, model access, and manage application resources. Administrators assign roles and resource permissions to support centralized control and delegated team administration. Each team has its own workspace, with its own policies, models, and logs, under central governance controls.
  • Multilayered, high-performance guardrails: FortiAIGate integrates Virtue AI’s high-accuracy, low-latency guardrails to inspect prompts and outputs for unsafe, harmful, or policy-violating content. These guardrails inspect prompts and responses across nine languages. Built-in data loss prevention (DLP) detects 35 types of sensitive data and then blocks or redacts it. Administrators can write custom detectors in natural language, specifying what to block and what to allow.
  • MCP protection: MCP tool scanning inspects tool lists, tool calls, and tool responses in both directions. It detects prompt injection, data leaks, and toxic content in agent traffic to and from MCP servers.
  • Compliance and policy: More than 50 predefined policies cover AI governance frameworks, data privacy laws, industry regulations, and security frameworks. Administrators add each one with a single click and customize the rules to fit the business. Workspace policies go through an approval workflow before taking effect. Logs of prompts, outputs, and tool calls routed through the gateway provide evidence for audits.
  • Model access and cost control: FortiAIGate tracks request costs based on provider pricing. Administrators set budgets for the organization, each workspace, project, user, and virtual key, and control which models each team can access. Budgets reset daily, weekly, or monthly, and FortiAIGate blocks requests when the spending limit is reached. Rate limits cap requests and tokens per minute. Guardrails are in place to block malicious prompts and denial-of-wallet attacks before requests reach the model.
  • LLM proxy and intelligent routing: FortiAIGate provides content-based routing, reverse proxy capabilities, and load balancing across model groups. Administrators can configure fallback rules to create an ordered recovery chain for a primary AI model. Content-aware routing routes routine tasks to lower-cost models and reserves premium models for work that requires them.

FortiAIGate and OCI Reference Architecture

Figure 1 illustrates how OCI Enterprise AI and FortiAIGate can work together to secure AI applications and agents. FortiAIGate can run on GPU-enabled Oracle Kubernetes Engine (OKE) worker nodes to inspect traffic to and from OCI Enterprise AI.

In this example, an AI chatbot routes requests through FortiAIGate to LLMs accessed through OCI Enterprise AI. The gateway runs in an OKE cluster within a virtual cloud network (VCN), where an ingress controller and load balancer manage incoming traffic. GPU-enabled worker nodes support security inspections. In a hub-and-spoke deployment, AI agents and applications can run in spoke VCNs connected to a hub VCN through an Oracle Dynamic Routing Gateway (DRG). The hub hosts security functions such as FortiAIGate, which can route inspected requests to OCI Enterprise AI, external providers, or services in other public clouds. Self-hosting keeps the gateway’s inspection and logs inside the organization’s environment. Where and how models are processed depends on the selected provider and deployment configuration, so organizations must choose routing and models that meet their data residency requirements.
Figure 1. FortiAIGate and OCI reference architecture

In this example, an AI chatbot routes requests through FortiAIGate to LLMs accessed through OCI Enterprise AI. The gateway runs in an OKE cluster within a virtual cloud network (VCN), where an ingress controller and load balancer manage incoming traffic. GPU-enabled worker nodes support security inspections.

In a hub-and-spoke deployment, AI agents and applications can run in spoke VCNs connected to a hub VCN through an Oracle Dynamic Routing Gateway (DRG). The hub hosts security functions such as FortiAIGate, which can route inspected requests to OCI Enterprise AI, external providers, or services in other public clouds.

Self-hosting keeps the gateway’s inspection and logs inside the organization’s environment. Where and how models are processed depends on the selected provider and deployment configuration, so organizations must choose routing and models that meet their data residency requirements.

Scaling Enterprise AI with Consistent Security

Autonomous AI agents can boost productivity, but their access to enterprise data and tools calls for security controls across the workflows they perform. Organizations need visibility into these activities, consistent policy enforcement, and safeguards to keep AI spending within approved budgets.

Together, OCI and Fortinet combine managed AI infrastructure with runtime inspection and cost controls. These capabilities help organizations scale AI applications and agents while safeguarding proprietary data and maintaining operational control.

Learn more about OCI Enterprise AI and explore Fortinet solutions on Oracle Cloud Infrastructure marketplace to help secure AI applications and agents, enforce policies, and control costs.