APPLE
Apple says it’s figured out how to improve its AI models, without ever needing to read your messages or emails.
Instead of collecting personal data, it uses a synthetic dataset that your device quietly compares to your recent messages (only if you’ve opted into analytics)
The device doesn’t send the actual data, just a signal about which fake example is the closest match.
Apple then uses the most commonly matched samples to fine-tune its AI features, like writing email summaries.
This is part of Apple’s wider focus on privacy, using a method called differential privacy to keep everything anonymous.
The approach is now being tested in beta versions of iOS 18.5, iPadOS 18.5, and macOS 15.5.
Here’s what you should know:
Devices compare fake data to real messages and only send anonymised signals
No personal data leaves the device
Testing is underway in upcoming beta software releases
Fake it, rate it, train it
While Apple’s AI rollout has had some delays and leadership changes, this latest update is a sign it’s pushing ahead, with privacy still front and centre.
Tech bros are somewhere sweating over this.