This preprint develops the Agentic Supply Chain Atlas: a public-evidence method for mapping what AI agents can touch. This preprint develops the Agentic Supply Chain Atlas: a public-evidence method for mapping what AI agents can touch. Rather than asking only which agents exist, it records delegated actions, tools and APIs, memory and retention claims, human approval gates, audit/provenance claims, MCP/OpenAPI/plugin/browser/RPA integrations, trust centers, system cards, model cards, and incident-policy evidence. The paper frames these facts as candidate regulatory evidence under the EU AI Act without claiming certification, conformity assessment, or legal compliance. Rows, mappings, and examples are candidate mappings based on public evidence. They are not legal conclusions, conformity assessments, certifications, or legal advice. Companion corpus and reproducibility bundle: https://doi.org/10.5281/zenodo.20489506. Deposited artifact SHA-256: 14f169e32477c2f55bdbf25a6c9401644b2fae940afcc35814735c03fb4ea497. This Zenodo record is used as an interim public manuscript/preprint route while OSF/MetaArXiv submission awaits user-provided OSF authentication.
Anton Sokolov (Mon,) studied this question.