Secure and govern AI apps and agent traffic
Stop AI attacks and data leaks, enforce usage policies, and keep AI spend within budget from one self-hosted AI gateway
Download Solution Brief Watch Demo
Overview
Safely move AI into production
FortiAIGate secures and governs AI at runtime from one control point. It stops AI attacks and data leaks, enforces usage policies, and keeps AI spend within budget.
- Keep your data under your controlPrompts, data, and logs stay on your infrastructure with self-hosting.
- Close blind spots across formatsGuardrails check text, images, audio, video, and code.
- Keep AI applications fastLow-latency guardrails protect users without slowing them down.
Outcomes
Secure AI, govern use, and cap spend
Runtime security
Stop harmful AI content in real time
FortiAIGate protects AI apps and agents at runtime. Guardrails block harmful prompts, outputs, and agent tool calls before they reach users or systems. They catch prompt injection, jailbreaks, and data leaks in Model Context Protocol (MCP) tool traffic. Very low latency keeps applications fast, and very low false positives keep analyst focus on real threats.
- Block prompt injection and jailbreaksInput guardrails inspect prompts and MCP tool responses before they reach the model. Hidden instructions stop before they steer your agents.
- Stop sensitive data leaksOutput guardrails inspect responses and tool calls. Data loss prevention (DLP) detects 35 types of personally identifiable information (PII) and confidential data, then redacts sensitive data in responses and blocks it in tool calls.
- Cover text, images, audio, video, and codeOne policy applies across five content types. Injections hidden in an image or a file get the same scrutiny as text.
Cost control
Prevent overspending on AI
FortiAIGate shows who spends what on AI and keeps AI spend within plan. Set budgets and rate limits for each organization, workspace, project, and user. FortiAIGate blocks requests past the cap. Intelligent routing saves premium models for the work that needs them.
- Set budgets by organization, workspace, project, and userGive each owner a budget and a spend limit. Your administrators can track AI cost by owner.
- Stop overspending with hard capsFortiAIGate blocks requests once spending exceeds a limit. Budgets reset daily, weekly, or monthly, and rate limits cap requests and tokens per minute.
- Route each request to the right modelIntelligent routing sends each request to the right model based on prompt content, the requested model, and HTTP headers, so routine work goes to lower-cost models. Blocking malicious prompts stops denial-of-wallet attacks before they consume tokens.
Governance and compliance
Govern teams and agent traffic, and prove compliance
Each team works in its own workspace, with its own models, budgets, and policies under central control. You see each user’s AI activity. Auditors get evidence for each prompt, output, and tool call.
- Give each team its own workspaceWorkspaces separate policies, models, budgets, and logs. Roles and virtual keys tie each request to its owner. The built-in GUI enables easy management with no Python required.
- Start from a governed baseline in one clickChoose from more than 50 one-click regulatory and framework detectors, customizable to meet specific requirements. Protections align with OWASP Top 10 lists for large language model (LLM), MCP, and agentic apps.
- See deeper into shadow AIFortiGate sends the AI traffic it detects to FortiAIGate, which identifies the user, the AI service, and the content, then blocks the request or masks sensitive data.
Deployment
Scale protection with your AI workloads
AI traffic grows as teams add apps, agents, and models. FortiAIGate grows with it on your own infrastructure. Easily add capacity as demand rises, keep latency low at high volume, and keep guardrails on each prompt and response.
- Add capacity as AI traffic growsFortiAIGate uses your existing Kubernetes ingress controller, and you add inference replicas and GPUs as AI traffic grows.
- Keep latency low at high volumeMulti-GPU acceleration handles proxy work as request volume climbs.
- Protect prompts and responses in both directionsGuardrails inspect each prompt before it reaches a model and each response before it returns to applications and agents. DLP also redacts PII before a request reaches the model.
How it works
See FortiAIGate in action
- Enforce AI usage policies and stop sensitive data exposure
Enforce AI usage policies and stop sensitive data exposure
You choose which detectors and guardrails run, set the action for each, and apply policies by workspace or across the organization. When a request carries a prompt injection or attempts data exfiltration, FortiAIGate blocks the activity and logs the attempt.
- Cap AI spend without slowing your teams
Cap AI spend without slowing your teams
Administrators set budgets and rate limits for each organization, workspace, project, and user, and FortiAIGate enforces them automatically. Requests stop at the cap and budgets reset on schedule, so spending stays predictable.
- Protect PII in every prompt, response, and tool call
Protect PII in every prompt, response, and tool call
Detectors find personal and confidential data in prompts and responses, across 35 data types in six categories. When a user enters PII into a chatbot, FortiAIGate can automatically block or redact it before it reaches the model, and safe AI work continues.
Enforce AI usage policies and stop sensitive data exposure
You choose which detectors and guardrails run, set the action for each, and apply policies by workspace or across the organization. When a request carries a prompt injection or attempts data exfiltration, FortiAIGate blocks the activity and logs the attempt.
Cap AI spend without slowing your teams
Administrators set budgets and rate limits for each organization, workspace, project, and user, and FortiAIGate enforces them automatically. Requests stop at the cap and budgets reset on schedule, so spending stays predictable.
Protect PII in every prompt, response, and tool call
Detectors find personal and confidential data in prompts and responses, across 35 data types in six categories. When a user enters PII into a chatbot, FortiAIGate can automatically block or redact it before it reaches the model, and safe AI work continues.
Models & specs
Deploy on your infrastructure
Deploy FortiAIGate without re-architecture and keep AI traffic on your infrastructure. FortiAIGate runs on Kubernetes with NVIDIA GPUs, on bare metal, KVM, VMware ESXi, Amazon EC2, or Amazon EKS. Apps and agents connect through OpenAI- or Anthropic-compatible APIs.
Resources
Solution Briefs
Blogs
Checklists
Data Sheets
Point of View
Videos
Protect PII in Every Prompt, Response, and Tool Call »
Detectors find personal and confidential data in prompts and responses, across 35 data types in six categories. When a user enters PII into a chatbot, FortiAIGate can automatically block or redact it before it reaches the model, and safe AI work continues.
Cap AI Spend Without Slowing Your Teams »
Administrators set budgets and rate limits for each organization, workspace, project, and user, and FortiAIGate enforces them automatically. Requests stop at the cap and budgets reset on schedule, so spending stays predictable.
Enforce AI Usage Policies and Stop Sensitive Data Exposure »
You choose which detectors and guardrails run, set the action for each, and apply policies by workspace or across the organization. When a request carries a prompt injection or attempts data exfiltration, FortiAIGate blocks the activity and logs the attempt.
FortiAIGate FAQs
What is FortiAIGate?
FortiAIGate is a self-hosted AI gateway for AI security and governance. It sits between your apps, agents, LLMs, and MCP tools. From one control point, it stops AI attacks and data leaks, enforces usage policy, and controls AI spend. It also extends visibility to shadow AI detected by FortiGate.
What does FortiAIGate protect against?
FortiAIGate protects AI apps, agents, LLMs, and MCP tool traffic. It blocks a range of threats, including prompt injection, jailbreaks, data leaks, malicious code generation, model poisoning, toxic content, and denial-of-wallet (DoW) attacks.
How is FortiAIGate different from firewalls and web application firewalls (WAFs)?
Firewalls, WAFs, and endpoint tools inspect IP addresses, signatures, and payloads. They do not read the meaning of prompts and responses. FortiAIGate reads AI traffic in context, between applications, agents, and models.
Does FortiAIGate support on-premises and cloud-based LLM deployments?
Yes. FortiAIGate is self-hosted on bare metal, KVM, VMware ESXi, Amazon EC2, or Amazon EKS. It connects your apps to public, SaaS, and self-hosted LLMs. Supported providers include OpenAI, Anthropic, Google Gemini, Amazon Cohere and Mistral, plus Microsoft Azure AI Foundry, and Amazon Bedrock.
How does FortiAIGate control AI costs?
FortiAIGate enables you to set budgets, spend limits, and rate limits by user, project, and team. FortiAIGate blocks requests past the cap and tracks the cost of each prompt and response.
How does FortiAIGate address compliance requirements?
Start from a governed baseline with more than 50 one-click compliance policies. They include the General Data Protection Regulation and the NIST AI Risk Management Framework. Organizations can customize each policy. Logs of each prompt, output, and tool call give your auditors evidence.
How does FortiAIGate work with FortiGate?
FortiGate controls employee access to AI apps at the network edge. FortiGate also sends the AI traffic it detects to FortiAIGate. FortiAIGate identifies the user, the AI service, and the content, then blocks the request or masks sensitive data. It can push policy updates back to FortiGate.
