Apple’s promise was simple: Personal data from iPhone users will never be used to train AI models. That is now over.
The greed for ever more intimate AI training data does not stop at Apple either: To further develop Siri AI and train its in-house language models, the iPhone manufacturer wants to use personal user data from Siri conversations for the first time – including audio recordings themselves. The company had previously categorically ruled out such use of user data for training AI models – clearly positioning itself against the competition.
Apple can “store your entire interactions with Siri and the dictation function, including the audio data and transcripts of your typed and spoken Siri requests, as well as Siri’s responses”, according to a privacy policy found in the latest pre-release version of iOS 27. The data will not be linked to the Apple Account, the company emphasizes. “Apple takes steps to ensure that personal data is not stored or used by Apple”. Some of this user data may be reviewed by unspecified “review personnel”. The audio material will only be listened to by Apple employees, it continues.
Opt-in when activating Siri AI
Apple is relying on an opt-in procedure: When activating Siri AI, the manufacturer asks for help in improving the chatbot, as developers with access to the release candidate of iOS 27 report. The update will be released to the public on Monday, September 14. You can either agree to this or press a visually inconspicuous “Not now” button. The dialog cannot be completely dismissed. Users can switch it off again after activation in “Settings > Analysis & Improvements”.
Videos by heise
Previously, the company had quietly revised its guidelines on “responsible AI”. The sentence “Your private personal data and interactions are never used to train our foundation models” previously ended with a period. Apple has now removed this and added the following passage: “unless you explicitly choose to help improve them”.
Apple Watch provides AI summary of conversations
Apple is thus not only backing away from its previous promise never to use user data for AI training, but also undermining the promise of its “Private Cloud Compute”: This is intended to work just as privately as local AI models, but offer the performance of cloud frontier models. User data will not be stored long-term or made accessible to third parties at any point, as Apple emphasizes. Both Siri AI and many Apple Intelligence functions rely on Private Cloud Compute. With iOS 27, Google servers are also part of Apple’s Private Cloud Compute for the first time.
The about-face comes at a most inopportune time: In addition to the iPhone Duo, the manufacturer also introduced two Apple Watch models this week, designed to listen in permanently. This is intended to allow the wearer to receive a conversation transcript of the preceding 15 seconds at any time – as well as a summary of all daily conversations as “Siri Recap”. According to Apple’s description, the latter also relies on Private Cloud Compute.
Several years ago, Apple experienced its biggest data protection scandal to date after employees of subcontractors reported on their daily work, which consisted of evaluating Siri audio recordings. The often unintentional Siri recordings contained intimate details and personal information, and the work was carried out under lax data protection regulations. Apple then temporarily stopped the audio evaluation of Siri recordings and brought it back with an opt-in.
Empfohlener redaktioneller Inhalt #
Mit Ihrer Zustimmung wird hier ein externer Podcast (Podigee GmbH) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden.
Damit können personenbezogene Daten an Drittplattformen (Podigee GmbH) übermittelt werden.
Mehr dazu in unserer
[Datenschutzerklärung](https://www.heise.de/Datenschutzerklaerung-der-Heise-Medien-GmbH-Co-KG-4860.html).
([lbe](mailto:lbe@mac-and-i.de))