Skip to main content
SAICA-KG

Taxonomies

Swiss Cheese Model for AI Safety: Multi-Layered Runtime Guardrail Taxonomy

Shamsujjoha, Lu, Zhao, Zhu (CSIRO Data61) · v2025-01 (v3) · first published 2024-08-05

Scope

A SLR-derived taxonomy of multi-layered runtime guardrails for foundation-model-based agents. Organized along three architectural dimensions (pipelines, artifacts, actions) plus quality attributes that guardrails should satisfy. Evoked as a "Swiss Cheese" defense-in-depth reference architecture for AI safety by design.

Source →License: arXiv preprint; CC-BY

Why it's ingested

The Swiss Cheese Model paper gives SAICA-KG's Tool ontology an external architectural anchor for defense-in-depth reasoning: Tool nodes can be mapped to (pipeline stage x artifact x action) triples and evaluated against the 14 quality attributes. Strong crosswalks: SC-PIPE-01/02/03 aligns directly with SAICA-KG's TemporalPhase partition; SC-ART-* overlaps with LocusOfControl; SC-ACT-* overlaps with ControlParadigm (Block -> prevention, Flag -> detection, Modify/Retry -> correction, Fall-back/Isolate -> recovery). This makes the Swiss Cheese paper a first-class validation anchor for SAICA-KG's axis choices.

Categories & crosswalks

Each row lists one external category and any SAICA-KG FailureModes that cross-walk into it.

Notes

Category labels derived from the arXiv v3 HTML; quality-attribute and action lists are canonical. Parent_id groupings (SC-QA, SC-PIPE, SC-ART, SC-ACT) are synthetic grouping anchors used for hierarchy — the underlying paper treats them as dimensions rather than named parents. Pending full-text verification against the final ICSA 2025 version, including the Rules, Sources, Modality, and Underlying-Models dimensions which are deferred to v0.2 expansion.

View raw YAML on GitHub →