al-ice.ai

CIS Extends Security Controls to AI Agents and MCP

Security by al-ice.ai Editorial

AI relevance: The Center for Internet Security has published the first framework that maps CIS Controls v8.1 — the de facto baseline for enterprise security — across all three layers of the agentic AI stack: LLMs, autonomous agents, and the Model Context Protocol.

Key Details

  • CIS released three AI companion guides in partnership with Astrix Security (non-human identity) and Cequence Security (API security): one for LLMs, one for AI agents, and one for Model Context Protocol (MCP).
  • The LLM guide addresses prompt injection risks, context window protection, and sensitive data handling during inference.
  • The agent guide covers safe tool execution, governed autonomy levels, and access control when agents interact with enterprise systems.
  • The MCP guide focuses on secure tool access, non-human identity management, and auditable interactions across the protocol layer.
  • CIS emphasizes that no single layer can secure the full AI system — controls must span input sanitization, context protection, tool validation, logging, and output review.
  • The guides are built on CIS Controls v8.1, meaning security teams already using the framework can extend existing programs rather than adopting a separate AI-specific standard.

Why It Matters

  • Most AI security guidance has been vendor-specific or academic. CIS Controls are widely adopted by enterprise security teams — this gives them a concrete path to govern AI deployments using familiar controls.
  • The three-layer model (LLM → Agent → MCP) maps directly to how organizations are deploying agentic AI, making it actionable rather than abstract.
  • The partnership with Astrix and Cequence signals that non-human identity and API security are now recognized as core AI security concerns.

What to Do

  • If your organization already implements CIS Controls v8.1, review the companion guides to identify gaps in your AI deployment posture.
  • Use the MCP guide to audit third-party MCP servers in your agent toolchain — especially around identity and access controls.
  • Map your existing agent governance policies to the controls outlined in the agent guide.

Sources