A message can be end-to-end encrypted and still be visible to an AI assistant after it appears on a phone screen. Signal is responding to that risk with new screen-capture protections intended to limit access by some built-in operating-system AI agents. The controls vary by platform, and Signal says they are not a complete privacy shield. Signal’s announcement describes the change as a response to AI features that can read what is displayed inside an app.

What Signal’s new protections do

On iOS 27, Signal says it will block screenshots in the app to prevent screen capture by AI features such as Siri’s Onscreen Awareness. On Android, the approach depends on the device and operating-system version: Signal says it will detect when an AI assistant tries to access the screen and turn on its Screen Security protection where it can identify that activity. Android versions differ, so this is not a guarantee that every device blocks every capture attempt.

CNET’s report says the protections are being enabled by default where supported. Signal says people who still want to take screenshots can switch off the protection in the app’s settings. That restores the convenience of screenshots, but also removes this layer of resistance to screen-reading assistants.

Signal previously blocked Microsoft Recall from capturing messages in its Windows app. The company says it cannot reliably apply the same protection on macOS. For people using macOS 27, its current workaround is to add Signal to the excluded-app list under System Settings > Siri.

Why encryption does not stop screen reading

End-to-end encryption protects messages as they travel between participants. A phone still has to decrypt a message to display it. An assistant with permission to inspect notifications or observe the screen may therefore access readable content on the device without breaking Signal’s encryption. The issue is what an operating-system agent can see after the app has opened the message.

This is not only a hypothetical design concern. In a separate investigation, Reuters reported that it worked with Signal and the AI Now Institute to reproduce access by Meta’s Muse agent. In those tests, Muse retrieved Signal message text through notification permissions on Android and through accessibility and screenshot tools on a Mac; the same request was blocked on iPhone. Meta told Reuters that Muse can access other apps only through permissions users explicitly grant and described the incidents as isolated.

A useful but incomplete layer

Signal is clear that screenshot controls do not close every route an AI agent might use. Its announcement warns that agents could potentially retrieve content already present on a device’s filesystem. The Reuters testing also illustrates that notification permissions can expose message text on Android, a route that is not the same as a screenshot.

That makes the feature a targeted reduction in risk, not a switch that disables AI access across a phone. It also leaves a broader question for app developers: can they prevent operating-system agents from accessing sensitive app data without breaking legitimate accessibility and device features?

Signal is asking Apple, Google and Microsoft for a well-documented way to opt apps out of AI data access, along with clearer technical documentation. Reuters and CyberScoop both describe the move as part of a growing tension between app-level privacy protections and AI features built into mobile operating systems.

For users, the practical takeaway is narrower than “AI can read every encrypted chat.” Signal’s change blocks some screen-capture paths on supported devices, while other permission-based access may remain possible. The message encryption can remain intact even as the privacy boundary around the unlocked device becomes more important.