A privacy line that once separated one phone maker from its rivals has quietly lost a clause. Recordings and typed requests that customers already agreed to share can now be reused to sharpen the models behind a voice assistant.
The change arrived with iOS 27. During setup for the rebuilt assistant, a screen asks whether the owner wants to help improve Siri and Apple Intelligence, offering Share Audio and Text or Not Now.
What the prompt actually changes
Turning the setting off later takes a journey through Privacy and Security, then Analytics and Improvements. That stops collection going forward without touching anything already stored.
Audio from Siri, Dictation and Translate was already gathered from people who opted in. The difference is purpose.
Material once used to tune recognition is now also routed into foundation model training, which is why everyone who said yes under the older wording gets asked again.
Apple says a random identifier generated on the device, refreshed several times an hour, stands in for the account. The company states the data is not tied to an Apple Account or email address, is not used to build a marketing profile and is not sold. A fresh support document lists the controls on every platform.
The caveat worth reading twice
Apple’s terms have long allowed audio to be reviewed by personnel after anonymization, and coverage of the reversal notes that transcripts may be read by reviewers whose employer is not disclosed. Agreeing to share is not the same as agreeing to be training material, which is exactly why the opt-in screen exists.
