CIS Releases Security Guides for LLMs, AI Agents and MCP
Summary
CIS, Astrix Security, and Cequence Security developed three Companion Guides that apply CIS Controls to securing LLMs, AI agents, and Model Context Protocol environments.
Key points
- The guides address three AI security layers: models and LLMs, autonomous agents, and the MCP protocol for connecting AI systems with tools and data.
- They extend existing CIS Controls and Safeguards for AI environments rather than introducing a separate framework.
- Guidance covers sanitizing inputs, protecting context and memory, constraining agent reasoning, authorizing tool requests, logging actions, and reviewing or minimizing outputs.
- The article identifies risks including sensitive-data leakage, unauthorized agent actions, persistent confidential data in memory, poisoned RAG pipelines, and unsafe MCP server capabilities.
- Astrix contributed expertise in agent, MCP, and non-human identity security; Cequence contributed enterprise application and API security expertise.
Article Details
- Event Type
- Publication of three CIS Companion Guides extending the CIS Controls to AI large language models, AI agents, and the Model Context Protocol (MCP).
- Impact
- The guides provide enterprises with layered cybersecurity guidance for AI systems, addressing model and data exposure, agent autonomy and tool use, identity and authorization, MCP access to tools and data, and auditing. The article describes security risks that AI deployments can introduce but does not report a specific incident or confirmed impact.
Vendors
Astrix SecurityTo address these challenges, the Center for Internet Security® (CIS®), Astrix Security, and Cequence Security partnered to develop actionable cybersecurity guidance tailored to AI environments.Center for Internet SecurityTo address these challenges, the Center for Internet Security® (CIS®), Astrix Security, and Cequence Security partnered to develop actionable cybersecurity guidance tailored to AI environments.Cequence SecurityTo address these challenges, the Center for Internet Security® (CIS®), Astrix Security, and Cequence Security partnered to develop actionable cybersecurity guidance tailored to AI environments.
Products
AI Agent Companion GuideThe result is three new CIS Companion Guides: the AI Large Language Models (LLM) Companion Guide, the AI Agent Companion Guide, and the Model Context Protocol (MCP) Companion Guide.AI Large Language Models (LLM) Companion GuideThe result is three new CIS Companion Guides: the AI Large Language Models (LLM) Companion Guide, the AI Agent Companion Guide, and the Model Context Protocol (MCP) Companion Guide.CIS Critical Security ControlsThe work extends the globally recognized CIS Critical Security Controls® (CIS Controls®) into environments where autonomous decision‑making, tool and API access, and automated threats introduce new risks.Model Context Protocol (MCP) Companion GuideThe result is three new CIS Companion Guides: the AI Large Language Models (LLM) Companion Guide, the AI Agent Companion Guide, and the Model Context Protocol (MCP) Companion Guide.