Apple revises App Store privacy labels to include on-device ML metrics

AI · 3 min read

Apple's App Store privacy labels will now include a dedicated section for on-device machine learning, requiring developers to disclose model types, data processed locally, and estimated storage and compute footprints. The new labels are designed to help users understand what intelligence runs on their device versus in the cloud.

The company provided guidance on estimating model size and compute, suggesting reporting thresholds and examples. For developers using third-party SDKs, Apple recommends including the SDK's ML disclosures or linking to the SDK vendor's privacy page.

The revision also introduces a 'local inference' badge on product pages to highlight apps that perform ML tasks exclusively on-device. Apple says the changes will roll out to the App Store over the coming months with tooling in App Store Connect to help generate the new label fields.