Apple has begun rolling out the most substantial redesign of Siri in the assistant’s history, and with iOS 27 the new Siri AI arrives as something closer to a system-wide reasoning layer than a voice command tool. It delivers contextual answers, searches across your installed apps, stitches together information from multiple sources and can act on the system on your behalf. That leap in capability comes with a matching leap in data access — far beyond anything the classic Siri ever required.
Access is currently gated behind a waiting list, but once activated, the system starts indexing notes, messages and, where developers have opted in, third-party app data. Apple worked with Google to train the language models powering Siri AI’s features. What follows is a walkthrough of every permission involved, what actually leaves your device, and how to dial back access without gutting the assistant’s usefulness.
- App Intents is the pivot: The new framework lets Apple Intelligence tools, Siri included, reach into data generated inside installed apps — by default, Apple’s own Notes and Messages first.
- Not everything stays local: Simple queries run on-device models, but heavier requests are routed to Apple’s Private Cloud Compute infrastructure, meaning data leaves the handset.
- Controls exist but are buried: Per-app indexing, Siri training opt-outs and even a rollback to Siri Classic are all available, though Apple has tucked them into non-obvious corners of Settings.
What Siri AI indexes, and why it differs from the old Siri
The core technical shift is the App Intents framework. It grants Apple Intelligence tools access to data created inside apps on your device. When you ask Siri AI something, it searches Notes, Messages and the rest of Apple’s app suite as a baseline. If an external developer has implemented App Intents support, Siri can search inside that app too.
Apple has also merged Spotlight and Siri under a single interface, widening the range of queries the system can handle. Previously Spotlight indexed files and app names; now Siri AI can pull semantic content from within apps rather than merely their titles or metadata.
The privacy implications are immediate. Days earlier, a columnist at Inc. wrote that Meta Muse appeared to know the contents of his messages. Apple responded with fresh restrictions on Full Disk Access in macOS to limit what agents can read. On iPhone, the equivalent vector isn’t Full Disk Access but App Intents — a more granular mechanism carrying the same underlying tension between utility and exposure.
Where your data goes: Private Cloud Compute and the limits of ‘on-device’
Siri AI does not process everything locally. It blends on-device models for straightforward requests with Apple’s Private Cloud Compute (PCC) infrastructure for queries demanding more computational horsepower. When PCC is engaged, data leaves the device and travels to Apple’s servers.
Apple states that it stores no data sent to PCC once a query is processed, and that not even Apple employees can view it. PCC’s technical architecture was designed around that goal and has undergone independent auditing. The distinction remains, however, that the data does exit the handset.
The Google collaboration adds another layer of complexity. Apple built much of its ecosystem on the promise that user data stays out of third-party hands, unlike its competitors. Partnering with Google — a company whose business model is built on data — to train the models behind Siri creates a legitimate tension, even though both firms contractually stipulate that Google has no access to production user data. OpenAI’s push to embed AI-driven smart cameras in smartphones, led by former Apple engineers, signals that rivals are heading toward the same deep system integration. Siri AI is the most sophisticated implementation today, but it will not be the only one.
Locking down indexing, training and the Siri version you run
The most practical control for most users is per-app indexing. Siri AI can reach any app that allows it by default, and switching it off app by app is straightforward:
- Open Settings.
- Scroll to Apps and tap the app you want to restrict.
- Tap Search.
- Turn off Show Content in Search.
With that disabled, Siri cannot extract information from that app even when asked directly. Disable indexing for Messages, for instance, and Siri will be unable to summarise a conversation on request. It is an explicit trade-off. Note that the Apps menu also houses an ‘Apple Intelligence & Siri’ section with options such as ‘Suggest App’ and ‘Learn from this App’. These govern when the system surfaces app suggestions in the search shade and how iOS tunes their timing to your habits — they carry no direct privacy implications for app data.
Apple does not collect your Siri interactions for training by default, but a rushed pass through the ‘Improve Siri’ screen during initial setup can leave it switched on. To verify: open Settings, tap Privacy & Security, scroll to Analytics & Improvements, and confirm that Improve Siri & Dictation is off.
Those who prefer the previous assistant can roll back, though Apple hides the switch inside Screen Time: open Settings, tap Screen Time, then Content & Privacy Restrictions, enable restrictions if needed, tap Siri, and under Allowed Siri Version select Siri Classic. To disable the assistant entirely, open Settings, tap Siri, scroll down and choose Turn Off Siri. Third-party voice apps remain usable with Siri off, and CarPlay voice commands continue to work under Siri Classic, which does not require Siri AI’s extended permissions.
Digital privacy remains a field where defaults outweigh optional configuration. The controls Apple built into iOS 27 are genuinely granular — but they only help users who know they exist. The steps above take under five minutes, and given how much of your personal correspondence now sits within reach of an assistant, that is time well spent.