NIST released a preliminary draft of its Cyber AI Profile, formally NIST IR 8596, a Cybersecurity Framework Profile for Artificial Intelligence, on December 16, 2025, and the draft's treatment of data provenance is its most consequential element for data engineering teams: it explicitly extends the kind of supply chain diligence that's become standard practice for software dependencies, knowing where your code came from, verifying it hasn't been tampered with, tracking its lineage, to the data feeding AI systems, treating compromised or opaque data as capable of undermining trust in an AI system just as effectively as a vulnerable software dependency can. This matters because data provenance has historically gotten far less rigorous treatment than software supply chain security in most organizations: teams that would never deploy an unverified third-party package without at least a cursory security review have often been comparatively casual about the lineage of training data, feature store inputs, or the datasets feeding retrieval-augmented generation pipelines, in part because tooling and standards for data provenance tracking have lagged well behind the software supply chain tooling ecosystem, SBOMs, dependency scanning, signed packages, that's matured over the past several years. A NIST framework that formally puts data provenance on the same footing as software provenance is likely to accelerate the development of comparable tooling and standards for tracking data lineage, similar to how earlier NIST and executive-order-driven software supply chain requirements pushed SBOM adoption from a niche practice to a widely expected baseline. Data engineering and platform teams building AI systems should treat this draft as an early signal to start documenting data lineage and provenance now, particularly for any pipelines feeding regulated or safety-relevant AI applications, rather than waiting for the framework to finalize and provenance requirements to become a compliance mandate.