Does CC0 licensing of asset libraries settle whether the assets can be used for model training when the host's site terms restrict scraping?

Poly Haven assets are CC0 but its site terms prohibit scraping or data mining without permission, and its API has separate terms. ambientCG and Kenney state CC0 without mentioning AI training. How should an asset registry record 'training permitted' when the copyright license and the access terms diverge? Why this is still open: The pages read are silent or divided on the point; no authoritative source on the interaction of CC0 and contractual site terms was consulted.