Andrew Dannenberger

I build AI security standards.

I am a member of the CIS Benchmark Development Team at the Center for Internet Security and the official Leader of the CIS AI Benchmarks Community. My work is focused strictly on building AI security standards: translating fast-moving AI risk into prescriptive, auditable recommendations that security teams can configure, verify, and audit.

My focus areas include the Model Context Protocol, LLM and agentic AI security, and security controls and Benchmarks.

I am the Author of the CIS MCP Server Benchmark v1.0.0, released September 2026, the first CIS Benchmark for Model Context Protocol servers: 55 prescriptive, vendor-neutral recommendations across 10 security domains, each with rationale, audit procedures, and remediation guidance.

I am also the Principal Co-author of the CIS Controls v8.1 Model Context Protocol (MCP) Companion Guide (April 2026), co-authored with Shreyans Mehta of Cequence, and a collaborator on the CIS Controls v8.1 AI/LLM Companion Guide and AI Agents Companion Guide.

As Leader of the CIS AI Benchmarks Community, I set the scope for new AI Benchmarks, coordinate community contributors through consensus review, and drive the publication roadmap for CIS AI security standards.

My standards work is grounded in hands-on AI engineering. I have developed MCP server concepts for an enterprise security platform, worked through agentic AI architectures and secure integration patterns, and built the interactive tools in my AI Security Lab. Writing the CIS MCP Server Benchmark meant turning that experience into audit and remediation procedures that teams can run against real MCP clients and servers. An earlier background in technical support for enterprise security products still keeps my recommendations tied to how systems actually behave in production.

My perspective is practitioner-first and implementation-aware. Good AI security guidance has to be specific enough to act on, not just a list of principles or a compliance checklist. That means understanding the protocols, the failure modes, and the organizational reality that security teams are operating in.

I've worked through MCP server concepts for a real enterprise platform, taught AI security to beginner and advanced audiences, authored the CIS MCP Server Benchmark, and co-authored a companion guide mapping MCP risk to all 18 Controls. When I write or present about AI security, I'm drawing on direct experience, not just research.

I also care that standards are usable. When guidance is too abstract or too hard to follow, teams route around it and end up with shadow AI: unapproved tools and unmanaged workflows. A recommendation that people can actually configure and audit reduces more real-world risk than a strict policy that quietly gets ignored.

Before transitioning into cybersecurity, I spent nearly a decade leading international adult education and training programs. At Raytheon Intelligence and Information Services, I served as Site Lead and Principal Training & Development Specialist on an international assignment, leading training programs for defense and government professional audiences in a high-pressure operational environment and leading a team of instructors while coordinating program delivery, quality, scheduling, and stakeholder communication.

Prior to that I delivered training and curriculum programs for adult learners across the United States and a range of international contexts, including professional, military, and higher-education audiences.

That experience, translating complex material for diverse adult learners, adapting to fast-changing environments, and making technical guidance accessible to people who need to act on it, directly informs how I write standards documentation, lead community consensus, and communicate AI security risk today.