AI Security Companion Guides Define Secure Enterprise AI, From Prompts To Protocols

The Center for Internet Security, Inc. (CIS), Astrix Security, and Cequence Security have jointly released three CIS Critical Security Controls Companion Guides to help enterprises secure rapidly evolving AI environments, extending the CIS Controls to large language models, autonomous agents, and Model Context Protocol integrations that introduce distinct security risks.

The AI LLM Companion Guide addresses security for large language models, with guidance on risks linked to prompts, context handling, and exposure of sensitive information. The AI Agent Companion Guide focuses on autonomous and semi-autonomous agents, highlighting safe tool execution, governed autonomy, and appropriate access to enterprise systems.

The MCP Companion Guide details protections for Model Context Protocol environments, emphasising secure tool access, management of Non-Human Identities such as API keys, service accounts, and OAuth tokens, and auditable interactions across the protocol layer, reflecting how modern AI systems are integrated into enterprise architectures.

As AI becomes embedded in production workflows, including copilots, autonomous task execution, and tool-integrated systems, security teams face risks such as data leakage, unbounded agent autonomy, credential misuse, and unsafe or inappropriate tool execution that traditional controls were not designed to handle.

"These guides reflect a shared effort to bring clarity to an area where organizations are seeking direction," said Curtis Dukes, Executive Vice President and General Manager of Security Best Practices at CIS. He said the collaboration translated the CIS Controls into concrete steps to secure AI systems across model, agent, and protocol layers.

Astrix contributed expertise in securing AI agents, MCP servers, and Non-Human Identities. "AI agents introduce a new operational surface that organizations must understand before they scale," said Jonathan Sander, Field CTO of Astrix Security, noting the focus on identity, authorisation, and execution risks aligned with enterprise practices.

Cequence contributed experience in securing enterprise applications, data, and APIs, shaping guidance around visibility, governance, and control over what AI systems can access and execute. Shreyans Mehta, CTO and Co-Founder of Cequence Security, said the partnership helped create guidelines for deploying agentic AI at large enterprises without sacrificing security, governance, or scale.

Together, the three Companion Guides adapt the CIS Controls to AI-driven architectures without introducing a new framework, offering prioritised recommendations that support responsible AI adoption across development, deployment, and operations, and covering the full AI security stack from model inputs to protocol-level access.

Overview of the CIS AI Companion Guides

CIS, Astrix, and Cequence will discuss key insights from the Companion Guides on 13 May at 1:00 p.m. ET during an event titled From Prompts to Protocols: The Security Blueprint for Enterprise AI, aimed at security teams, developers, and AI practitioners.