iOS 27 Adds Opt-In Setting Letting Users Share Data to Train Apple's Foundation Models

Apple shipped iOS 27 to the public today alongside the general release of Siri AI, introducing a new opt-in setting that lets users share audio and text interactions to help train Apple's Foundation Models, according to a report from 9to5Mac.

What changed

Previously, Apple asked users during setup whether they wanted to help improve Siri and Dictation by sharing audio and transcripts. With iOS 27, that prompt has been replaced by a new screen titled "Improve Siri & Apple Intelligence," offering two choices: "Share Audio and Text" or "Not Now."

Apple's own description of the setting reads: "Help improve foundation models and Siri by allowing Apple to collect and use your text and audio interactions with Siri, Dictation, and Translate."

Apple says the underlying data collected has not changed. What's different is how Apple uses it — the same audio and text data will now also feed into training and refining Apple's Foundation Models, in addition to improving Siri and Dictation as before.

Users who had already opted in under the old setting are not grandfathered in. They will be prompted again when they enable Siri AI, giving them a chance to reconsider now that their data may be used for foundation model training.

Privacy details

According to Apple, Siri audio data, transcripts, and related data are tied to a "random, device-generated identifier that rotates multiple times per hour" rather than to a user's Apple Account or email address. Apple states it does not use this data to build marketing profiles or sell it to third parties.

Users can review or change their choice at any time in Settings > Privacy & Security > Analytics & Improvements > Improve Siri & Apple Intelligence. Apple has also published a new support document detailing how these settings work across its platforms.

What this means

This is not a new model release — it's a data-collection policy change tied to the public rollout of Siri AI in iOS 27. The significance lies in Apple explicitly linking user opt-in data to foundation model training for the first time, a step it had avoided with earlier Apple Intelligence features that relied more heavily on synthetic data and third-party licensing.

Re-prompting existing opted-in users is a notable move: it signals Apple treats foundation-model training as a materially different use case requiring fresh consent, not just a continuation of prior Siri improvement programs. The rotating device identifier and no-marketing-profile claims are Apple's standard privacy playbook, but they remain unverified by independent audit — outside researchers have not confirmed how anonymization is implemented at scale.

For developers and enterprises watching Apple's on-device AI strategy, this suggests Apple is now building a first-party data pipeline for its foundation models, similar in spirit to how OpenAI and Google use opt-in user interactions to refine their models, but wrapped in Apple's stricter marketing-and-privacy positioning.