Cequence adds AI Gateway controls for agentic zero trust

Cequence has introduced new functions in its AI Gateway and updated its Agent Personas system. The release brings model, API and tool controls together under what it calls Agentic Zero Trust.

The update adds AI Discovery, API Registry, LLM Registry and Skill Registry. It also expands Agent Personas so an AI agent's assigned role determines which models, tools and services it can use.

Cequence is addressing a problem that has emerged as companies roll out AI agents across departments such as finance, marketing, human resources and operations. Security teams often have to review each new use case manually, creating delays and leaving governance split across separate systems for APIs, models and tools.

Under the new approach, an agent's identity is linked to a defined job description. That description can specify approved large language models, permitted APIs, available tools and relevant guardrails, with those limits enforced through policy rather than case-by-case approval.

Unified controls

The release is designed to cover the main external channels used by AI agents. In Cequence's framework, MCP governs how agents discover and use tools, the LLM Registry governs calls to and from language models, and the API Registry governs access to back-end services and data.

Agent Personas sit across those layers by binding the agent to a single role-based identity. Cequence argues that this prevents agents from operating beyond their intended remit, even if they encounter exposed credentials or vulnerabilities elsewhere in a system.

The announcement comes amid wider concern over how autonomous software agents are controlled once connected to enterprise applications and data. Companies have adopted scanners, gateways and prompt filters, but many security leaders still lack a single record of which agents are in use, what they can access and how they can be shut down if controls are breached.

Cequence also cited a recent incident disclosed by OpenAI in which two models escaped a sandboxed evaluation environment, crossed the open internet and breached Hugging Face production infrastructure. It used that case to argue that sandboxing alone does not constrain an agent if no policy binds it to a specific job.

"Most vendors look at agent governance and build another approval queue. We looked at it and built the persona instead," said Shreyans Mehta, Chief Technology Officer and Co-Founder, Cequence. "An agent's job should automatically determine what it can touch, without relying on a security team to manually map policy by hand every time someone wants a new use case. That's what makes broad adoption safe and scalable. The agent gets exactly what its job requires, and nothing more."

What is new

AI Discovery is intended to identify agents, LLM providers and MCP servers already operating inside an organisation by drawing on existing SIEM logs, including systems that did not go through a formal approval process.

According to Cequence, API Registry allows agents to call approved APIs without holding the underlying credentials. Instead, agents authenticate through a single AI Gateway access key, either through a web-based invocation tool or proxied endpoints.

Skill Registry is aimed at security and platform teams that want a pre-approved set of reusable functions for agents. Once a tool or workflow is vetted, it can be reused across different agent deployments without repeating the same review from scratch.

LLM Registry extends that logic to model access. Cequence said it brokers credentials across major LLM providers so agents do not hold a provider API key directly, while built-in data loss prevention checks prompts and responses for blocked content, including encoded payloads and non-approved Unicode characters intended to evade filters.

The registry can also apply model rules at team level, such as steering routine work to lower-cost models while reserving more advanced models for engineering tasks. It also provides token-level usage visibility, plus rate and spending controls linked to the persona behind each request.

Policy mapping

A central part of the release is the claim that policy enforcement can now be automated because the tools, APIs and skills available to an agent are formally catalogued. Without that catalogue, policy decisions have often depended on manual judgment.

"Automatic policy mapping was not possible until now, because there was nothing consistent for a policy engine to reason over," said Abraham Jeevagunta, Vice President of AI Products, Cequence. "Before API Registry and Skill Registry, every tool and API a persona could be bound to was uncatalogued, so mapping policy to persona was a manual judgment call every time. Now, that record exists and the policy engine can read it directly. It is what lets a business user stand up a correctly governed agent without ever touching AI Gateway's policy model themselves."

The new functions are available as part of AI Gateway for existing Cequence customers. Cequence said its platform currently protects more than 10 billion daily API interactions and 4 billion user accounts.