Agentic AI Security — Reference Library

The public card catalogue for The Multiverse School's agentic-AI-security curriculum. Everything below links to primary sources; the numbers are verified source counts.

Start here — the interactive map

▶ Interactive framework map — one zooming canvas of the whole field: the three levels and the meta-map, navigable by camera (Tour or Explore).

Handouts (three levels)

The teaching handouts. These require a school login.

• Level 1 — Agentic SDLC + AI User — how we ship agents safely: user & abuse cases, secure-SDLC canon, threat modeling, agentic build patterns.
• Level 2 — AI Alignment & Governance — alignment, standards, and safety cases.
• Meta-Map — When & Where — which framework fires at which lifecycle phase.

Reference collections (public, citation-verified)

The source-of-truth bibliographies underneath the levels. 912 sources, every one linked.

• Level 3 — Research Frameworks — 312 sources across 14 clusters: the agentic-security research frontier.
• Deep Dive — Anthropic · Apollo · METR — 161 sources: the three bodies of work the course leans on hardest.
• Papergraph — Roots & Adjacent Work — 261 sources an org-shaped reading list misses: MIRI, ARC, Redwood (the AI-control agenda), the OpenAI/Distill circuits lineage, NYU, Epoch, Goodfire.
• AI-Assisted Hacking — Case Studies — 129 documented cases of AI used offensively (and turned back on the problem), each with a defensive lesson.
• Foundational Agenda Papers — 49 field-defining research agendas (the Concrete Problems / Foundational Challenges genre).

Compiled for The Multiverse School. Every framework, paper, and case links to its primary source — follow the links, don't trust the summaries.