cff-version: 1.2.0 message: "If you use this dataset, cite it as below." title: "Failures in Agentic AI: a five-dimension taxonomy and corpus of reported failures" type: dataset authors: - family-names: Aldahmani given-names: Maryam affiliation: United Arab Emirates University - family-names: Alhosani given-names: Shama affiliation: United Arab Emirates University year: 2026 license: CC-BY-4.0 url: "https://huggingface.co/datasets/agentic-failure-study/failures-in-agentic-ai" repository-code: "https://github.com/agentic-failure-study/agentic-failure-study" keywords: - agentic AI - multi-agent systems - failure taxonomy - empirical software engineering abstract: >- Labelled corpus of publicly reported failures of agentic and multi-agent applications, coded on five independent dimensions with per-record evidence spans and an audit trail of the rules that fired.