Back to articles

Signal Moves to Limit Mobile AI Access to Chat Screens

3 min read

Introduction

End-to-end encryption does not mean messages are inaccessible at every stage. Signal says AI assistants built into mobile operating systems may read chat content from the screen, and it plans to limit their ability to capture or store Signal data. The issue is not the encryption protocol itself, but what operating systems and their AI features can access after a message appears on a device.

Key points

  • Screen capture restrictions: Signal says it will block screenshots inside the app on iOS 27 and take similar steps on Android versions where it can detect an AI assistant trying to read the screen. The company previously used screenshot blocking to keep Microsoft Recall from capturing Signal messages by default on Windows 11.
  • Users can turn protection off: Signal says users can allow screenshots in the app’s settings, but doing so may expose their own messages and those of their contacts to operating-system AI. Android versions vary by manufacturer, and Signal says it cannot reliably detect the behavior on every device.
  • The fix has limits: Blocking screenshots does not prevent AI from retrieving data already on a device through the filesystem. Notifications, system interfaces, and accessibility permissions may also provide other paths. Signal says it cannot currently implement the same protections on macOS.
  • A call for platform tools: Signal wants Apple, Google, and Microsoft to provide documented, straightforward app-level controls to deny AI access to sensitive information—including screen content, files, and notifications—without compromising an app’s core functions.

Why it matters

Signal argues that end-to-end encryption protects communications at the protocol and transmission layers, but an operating-system agent that can read the screen may still capture chat content and pass it to a local database or remote service. Once the data leaves the app’s control, Signal has less ability to maintain its privacy boundary. In practice, an app’s security promises also depend on how the operating system it runs on handles AI permissions.

The situation highlights a challenge for developers: system-level AI can act across apps, but developers lack a consistent, well-documented way to refuse access. Signal says it is in discussions with some platform providers, but describes screenshot blocking as a limited safeguard, not a complete solution. For users, the choice between convenient AI features and limiting exposure of sensitive content may need to be made more explicit. For the industry, enabling AI agents to work without crossing application privacy boundaries is becoming a significant operating-system design question.

Web Search

Comments

Checking sign-in status...

Loading comments...

Related articles