Apple Changes AI Training Data Wording, Introduces Explicit Opt-In
Apple now allows private interactions to be used to improve foundation models, but only with the user’s explicit, voluntary consent.
Opt-in Apple Intelligence is a new exception that Apple added to its statements on training data on September 9, 2026. The company says it does not use users’ private personal data and interactions to train its foundation models unless a user explicitly chooses to help improve them.
This is a change in communication, not a claim that Apple will generally or without consent train AI on user data. Under the current wording, voluntary user enrollment is required for such data to be used.
What opt-in Apple Intelligence includes
Apple added the same exception to older technical articles covering Apple Foundation Models. The materials list an update dated September 9, 2026.
On its privacy pages, the company says that after voluntary participation, it may store and review audio recordings from Siri and Dictation. This may also include interaction history stored in the Siri app. Users can opt out of the program at any time.
Apple distinguishes this mode from the Device Analytics setting. For device analytics, it refers to techniques for collecting aggregate trends without collecting individual user content. Sharing data to improve Siri and Apple Intelligence therefore cannot automatically be equated with standard diagnostic analytics.
Why the meaning of privacy settings is changing
Apple has long presented Apple Intelligence as relying on on-device processing and the Private Cloud Compute system. The new wording clarifies that, with explicit consent, there may be a separate way for users to contribute their own interactions to improving the company’s systems.
Users should therefore distinguish between the service’s default operation, aggregate analytics, and voluntary data sharing to improve models. According to Apple’s materials, it also uses licensed, publicly available, and synthetic data; the new wording specifically concerns private data and interactions provided through opt-in.
Details about data processing are missing
Apple has not yet publicly described in detail the full technical process, scope, or specific retention periods for data that will be used directly to train foundation models after voluntary participation. It is also not known how many users have joined the program or what measurable benefit this data will provide for future models.
The exact wording of consent screens, their availability in individual regions and system versions, will need to be monitored. Any technical information about anonymization, human review, retention, and deletion of voluntarily provided data will also be relevant.
Sources
- Apple — Datasets used for Apple's generative AI systems and services – The official page dated September 9, 2026 confirms an explicit exception: private data and interactions may be used to train foundation models only if the user chooses to help with improvements.
- Apple Machine Learning Research — Introducing the Third Generation of Apple’s Foundation Models – The technical article was updated on September 9, 2026, to reflect the possibility for users to opt in to improving Siri and Apple Intelligence.
- Apple Privacy — Features – Apple says that with voluntary participation, it may store and review Siri and Dictation audio and interaction history in Siri; users can opt out.
- Apple Support — Applebot model training and individual privacy rights – Apple Support independently repeats the new opt-in rule and describes other training data sources, including licensed, publicly available, and synthetic data.
Verified and updated: 09/13/2026 06:24
