Auditors request model cards, sanctioned-tool lists, and sovereignty attestations for city AI systems. What are these?

Auditors of AI programs commonly request artifacts such as model cards, sanctioned-tool inventories, and data-sovereignty attestations. Providing those as compliance evidence for municipal AI systems is correct: they show what models do, which tools are approved, and where resident data and derived stores may live. Evidence artifacts support findings; they are not vendor selection scorecards supporting audits of municipal AI acquisitions, and they are not fine-tune evaluation harnesses supporting audits of municipal AI deployments. Treating them as vendor selection scorecards supporting audits of municipal AI acquisitions fails by concept: auditors use these artifacts to test design and operation of AI governance, not to review vendor acquisition justifications. Checkpoint lineage manifests supporting audits of municipal AI operations fails by relevance: auditors requested model cards, sanctioned-tool lists, and sovereignty attestations as compliance evidence, not checkpoint lineage manifests. Fine-tune evaluation harnesses supporting audits of municipal AI deployments fails differently: these test tuned-model performance for deployments, while the requested artifacts are compliance evidence that models, tools, and sovereignty controls exist and operate, not evaluation results.

Exam caveat: model cards, sanctioned-tool lists, and sovereignty attestations are compliance evidence artifacts—retain and present them; do not confuse them with vendor selection scorecards or with fine-tune evaluation harnesses.

Operational check: maintain current model cards, publish and enforce the sanctioned-tool list, keep sovereignty attestations aligned to actual hosting regions, and produce the packet promptly when municipal AI auditors request it.