How to Limit What Siri AI Can Access in iOS 27

According to Thorin Klosowski of EFF Deeplinks, Apple’s new OS, iOS 27, rebuilds the voice assistant Siri around an AI-centric design. It shifts from the traditional invocation methods to a standalone app, supporting conversational responses and cross-app summarization of on-device information. In exchange for enhanced capabilities, the expanded scope of access and decisions about external transmission have become less visible to users. The article outlines ways of thinking about narrowing the scope of access and settings to review.

Siri AI Overhaul and Integration Points

The new Siri is positioned differently from the conventional assistant that responded to short voice commands. It draws key points across a user’s notes, messages, emails, and more. It retains conversational context and answers follow-up questions in sequence. Invocation has also changed, operating as a standalone app in addition to voice- or button-based activation.

This change extends the context understanding covered in Apple Revamps Siri AI with Personal Context Feature. It is designed to improve answer accuracy based on personal information on the device. While convenience increases, it is harder to tell what information was used as the basis. Users must therefore decide the scope of access they permit.

Among competing voice assistants, as New Google Home Speaker Falls Short of 6-Year-Old Nest Audio on Sound Quality shows, factors other than functionality become points of comparison. For Siri AI, the criterion is not sound quality but how information is handled. Its distinctive feature is the deep coupling between the device and the compute infrastructure. Operation with awareness of where processing occurs is required.

How Search Changes with Spotlight Integration

In iOS 27, the touchpoints between Siri and Spotlight have been unified. Pulling down on the screen to find apps or contacts now directly invokes Siri. The visual distinction between the search box and the conversation window has blurred. Users are less likely to notice the boundary between search and generated responses.

By default, content from Apple-made apps becomes subject to answer generation. This includes notes, messages, emails, and more. Relevant portions are extracted according to the question and presented as summaries. It is a structure in which accumulated on-device data directly serves as a knowledge source.

This integration simplifies operation. At the same time, what was intended as a simple on-device search can trigger AI processing. To avoid unintended access, search targets and generation targets must be considered separately. Managing targets in the settings screen is a prerequisite.

The Opaque Boundary Between On-Device

Processing and PCC

Siri AI processing may be completed on-device or executed on Private Cloud Compute. PCC is Apple’s compute infrastructure, which processes information sent from the device. According to the company, it is designed not to retain information after processing. According to Thorin Klosowski of EFF Deeplinks, there is no direct on-screen indication of where processing takes place.

When a question is entered, it is impossible to tell whether it was processed on-device or sent externally. In some cases, checking the Apple Intelligence Report afterward can trace records of transmission. But it does not provide material for immediate judgment. Opacity remains in everyday use.

PCC is described as designed so that Apple cannot view or store information. Even so, the fact that information leaves the device does not change. Encrypted storage and external transmission are separate issues. Settings that reduce transmission itself are important.

Optimization of on-device processing is, as Linux 7.4 Improves Apple Silicon Audio Using Shared GPIO shows, an area advanced through tight coupling of hardware and OS. For Siri AI as well, device performance affects how processing is distributed. Not all requests are completed on-device. Users need to define the scope of access on the assumption of external transmission.

Third-Party Apps and How Onscreen Awareness Works

Handling of third-party apps depends on developer support. Content can be referenced only when developers have implemented integration for Siri AI. Unsupported apps are generally out of scope. An exception is capture via onscreen awareness.

Onscreen awareness is a method of sharing displayed content as an image. For example, even if Signal does not support integration, content shown on screen may still be in scope. The article cites a conversation with Bill as an example. While a summary of photos can be obtained in Apple’s Messages, no answer may be available in an unsupported messaging app.

Expanded integration improves convenience. At the same time, it leaves room for unexpected apps to be added to the scope of access. Newly installed apps and post-update permission changes should be checked. Users need to compare developers’ stated support with actual settings.

Settings to Limit Siri’s Scope of Access

The basis for narrowing access is per-app exclusion. Disable use in search targets, use in suggestions, and use in learning individually. Even for Apple-made apps, exclude them if they contain sensitive work or medical information. For third-party apps, review enabled integrations and disable unnecessary ones.

Treat onscreen awareness as an independent item. If you do not want capture via sharing, disable the feature. Because primary information on display directly becomes subject to processing, the impact is significant. Judge it with the understanding that it behaves like screen sharing.

Review how Siri itself is invoked. Limit voice invocation, button activation, and invocation via Spotlight according to use. Fewer invocation paths reduce accidental triggers and unintended transmission. Restricting invocation is effective on shared or work devices.

Make it a habit to check the Apple Intelligence Report. It allows retrospective confirmation of whether transmission to PCC occurred. Trends in the records help identify frequently referenced apps. They provide material for prioritizing exclusion settings.

Tension with Encryption Settings and

Direction for Countermeasures

Advanced Data Protection is a feature that applies end-to-end encryption to much of the information stored in iCloud. Encrypted storage on-device and transmission to PCC are different in nature. Even if transmission is designed to be unviewable, the fact that information leaves the device does not change. Evaluation criteria for risk need to be switched.

Attention is also needed to the nuance of the word “private.” Private in PCC refers to a state designed so that Apple cannot view or store information. It is not synonymous with encrypted communication channels or storage. It is important to reframe it from the perspective of whether transmission outside the device occurs.

Countermeasures boil down to minimizing access targets and reducing opportunities for transmission. Exclude sensitive apps, limit onscreen awareness, and narrow invocation paths. One option is to limit use to requests that stay within on-device processing. For organizational use, bulk restrictions via device management are worth considering.

Editorial Opinion

On short-term impact. Over the next three to six months, limiting Siri AI’s scope of access will become standard practice in enterprise device management. More organizations will formalize procedures for exclusion settings and disabling onscreen awareness. Reviewing the Apple Intelligence Report will be added to audit items.

On the long-term view. In one to three years, disclosure obligations for on-device processing versus external computation will become a policy issue. Designs without explicit indication of processing location will be at a disadvantage in procurement requirements. Developers’ integration implementations will be screened, and transparency of support will become a competitive factor.

Questions from the editorial team. Where is the threshold for accepting off-device transmission for the sake of convenience. Can image-based capture such as onscreen awareness be treated on par with normal integration. Is a design that leaves to users the decision to hand encrypted stored information to external computation sufficient. These issues need to be examined.

References

• “How to Limit What Apple’s New Siri AI Can Access in iOS 27”, by Thorin Klosowski — EFF Deeplinks, 2026-09-18T21:55:16.000Z (CC BY)
• Source URL: https://www.eff.org/deeplinks/2026/09/how-limit-what-apples-new-siri-ai-can-access-ios-27