Apple says raw audio from Siri Audio Intelligence stays inaccessible to it

Apple's announcement at the iPhone Duo launch event couples new AI-driven listening features with a strong privacy stance. The features—Siri Recap, Live Rewind, Sound Recognition, and Music Recognition—depend on continuous audio processing, which could raise concerns about user surveillance. By introducing the Secure Exclave on the S11 chip, Apple aims to prove that raw audio never leaves protected hardware and is not accessible to itself or other software.

The design is a deliberate architectural choice: the microphone feeds an audio stream that is continuously overwritten, with no transcription or file creation. Encryption is applied if data transfers to an iPhone, and user text is protected with end-to-end encryption under certain conditions. This approach contrasts with cloud-based processing, suggesting Apple sees privacy as a key selling point for the Apple Watch Series 12 and Ultra 4.

The stakes hinge on whether the Secure Exclave's isolation holds in practice. If vulnerabilities are found, it would undermine trust in Apple's core privacy promise. The claim that even Apple cannot access the audio is absolute; the test is whether third-party security researchers can validate this in real-world conditions over time.