F5 expands AI Gateway to curb costs & secure agents

F5 has expanded its AI Gateway and integrated it into the F5 AI Security Platform, targeting organisations managing growing volumes of AI inference traffic.

The updated product combines policy enforcement, model access controls and monitoring for AI models, agents and tools in a single system. F5 is positioning the gateway as a central point for managing the cost, governance and security issues that come with broader use of generative AI in business operations.

Demand for that level of control is rising as companies move from trials to regular deployment. F5 cited its 2026 State of Application Strategy Report, which found that 77 per cent of organisations now see inference as their main AI activity, while the average organisation manages seven AI models.

That shift matters because inference creates a steady stream of requests, each with spending, performance and security implications. As more business teams adopt AI tools and more software agents connect to data sources and application programming interfaces, companies face a more complex task in deciding who can use which model, under what rules and at what cost.

According to F5, the gateway brings together three elements: a Model Gateway for access and cost management, an MCP Gateway for governing how agents connect to tools, and AI Guardrails for inspecting prompts and responses. Budgets, routing policies and access controls can be set centrally and applied across different environments.

Cost focus

A key part of the update is closer scrutiny of token use, reflecting a wider industry focus on the economics of AI workloads. Tokens are the units used to measure text handled by large language models and are closely tied to pricing in commercial AI services.

F5 said the Model Gateway can attribute token consumption by provider, model, team and user. It also lets teams set budgets that are enforced as spending occurs rather than after usage has already been billed.

The system can also route requests between models based on cost and policy, while using semantic caching and GPU-aware load balancing to reduce unnecessary spending. F5 said the setup is designed to cut token spending by up to 60 per cent without requiring application changes.

Agent controls

The product also addresses management of AI agents, which are increasingly being used to automate tasks by calling external tools and data sources. As their use expands, companies face questions about how to authorise access and track what those agents do on behalf of users or departments.

F5 said its MCP Gateway introduces access controls that limit agents to approved resources, including APIs, data sources and retrieval-augmented generation systems. It also keeps an audit trail showing what was accessed, when, which agent made the request and on whose behalf.

F5 added that a registry of approved MCP servers is intended to give development teams a single source of reference for available tools. That reflects a broader push among infrastructure suppliers to bring more formal control to emerging agent-based software patterns.

Security layer

Another part of the update focuses on inspecting prompts and responses before they are sent to or returned from AI models. Organisations in regulated sectors are increasingly concerned about the risk of sensitive information being exposed to external models or malicious prompts bypassing safeguards.

F5 said its AI Guardrails can redact sensitive data before it reaches a model, block prompt injection and jailbreak attempts, and stop requests when they cannot be evaluated safely. The gateway also includes audit trails, security information and event management export functions, and data residency controls.

Those controls are intended to support internal governance and compliance work in sectors where health records, personal information and intellectual property may flow through AI systems. F5 said the system aligns with SOC 2, ISO and HIPAA frameworks.

The broader F5 AI Security Platform was introduced earlier as a framework for visibility, governance, testing and runtime protection across AI applications and the interfaces connected to them. By placing the AI Gateway inside that platform, F5 is tying those oversight functions more closely to live operational traffic.

That places F5 in a market where suppliers are trying to establish themselves as intermediaries between enterprises and the growing mix of third-party and internal AI models. The company cited Gartner research describing AI gateways as an increasingly important part of AI infrastructure for large enterprises seeking controlled access to models and MCP servers.

"We're watching enterprises race to deploy AI while struggling to control it," said Kunal Anand, chief product officer at F5.

"Every AI request carries economic, security and governance implications, yet most organisations are relying on fragmented tools that address only part of the problem. The result is rising costs, increased risk and operational complexity. F5 AI Gateway, integrated into the F5 AI Security Platform, provides a single control point for managing AI across models, clouds, agents and applications. We believe every enterprise will need an intelligent control layer for AI. F5 is building that foundation, helping customers accelerate innovation while maintaining visibility, security and control," Anand said.