Signal is changing how its messaging app handles screen capture to prevent AI assistants from accessing private conversations. The new restrictions affect iPhones running iOS 27 and supported Android devices, with different behaviour depending on the operating system.
The decision addresses a privacy problem that encryption alone cannot solve. Signal protects messages while they travel between devices, but an AI assistant with permission to examine the display may still collect information after a conversation has been opened.
That creates another possible destination for sensitive messages, outside the messaging service’s control. Signal president Meredith Whittaker raised the concern in a 20 November announcement, arguing that operating-system AI features are making it harder for independent apps to protect their users.
What changes for Signal users?
On iPhones, Signal 8.37 introduces a Block Siri AI option for devices running iOS 27 or later. It is enabled automatically and prevents screen capture within the messaging app.
Users who need to save a screenshot can temporarily suspend the restriction for ten minutes or disable it entirely. The option is available under Settings, Privacy, Block Siri AI.
Android takes a different approach. Beginning with Signal 8.35.1, the app can activate its Screen security feature when it detects an AI assistant attempting to inspect its contents.
This means screenshot protection does not necessarily start immediately on every Android phone. Users can enable it themselves under Settings, Privacy, App security, Screen security.
Signal also provides an explanation when the Android restriction is activated automatically. If users subsequently disable it, the app will not automatically restore the protection for 90 days.
These controls are intended to prevent assistants such as Gemini and Siri AI from obtaining information directly from Signal’s display. They do not change the encryption protecting messages or the people authorised to receive them.
Why screenshots are only part of the problem
Blocking screen capture addresses one method of collecting private information, but it cannot prevent every form of AI access.
An assistant with broader permissions might obtain information through another part of the operating system, rather than by examining an open conversation. Signal therefore considers its new safeguards a partial defence.
The company also faces a particular problem on macOS, where it cannot reliably prevent screenshots using the tools currently available to developers.
For people using Signal on macOS 27, its recommendation is to exclude the messaging app from Siri AI access through System Settings, Siri, Excluded Apps.
Signal has previously confronted a similar issue on Windows 11. In 2025, it introduced screenshot restrictions to prevent Microsoft’s Recall feature from preserving copies of conversations in its searchable activity history.
The concern extends to disappearing messages. Removing a message from Signal does not necessarily erase information that another system has already captured.
Signal wants stronger privacy controls from Apple and Google
The latest changes also expose a wider disagreement over how much access AI assistants should have to third-party applications.
Signal argues that developers need dependable controls over whether their apps can be examined by system-level AI. It has asked Apple, Google and Microsoft to provide clearer technical documentation and mechanisms for limiting access without disabling ordinary application features.
The company says it has been investigating how AI assistants interact with mobile apps because existing developer information does not adequately explain those interactions.
Signal reports that Apple and Google have engaged with it on the issue, although it has not announced an agreement or a timetable for broader protections.
For users, the immediate consequence is straightforward: protecting conversations from AI screen reading may also mean giving up the ability to capture screenshots. Signal is leaving that choice available while making privacy protection the default wherever its current implementation allows it.



