Signal, the encrypted messaging service, said on Friday, 20 November 2026, that it is restricting the artificial intelligence assistants built into phone operating systems from reading and storing users' chats. In a blog post by Signal president Meredith Whittaker, the non-profit said it now blocks screenshots inside its app on Apple's iOS 27 and switches on its "Screen security" setting on Android when it detects an AI assistant trying to read the screen.
Whittaker described the spread of system-level AI agents as "an existential threat to Signal" and to privacy at the level of individual apps. Her argument is that once a message appears on screen, an assistant that can see the screen can copy it out of Signal's encrypted environment, and Signal has no control over what happens to it after that.
What exactly is Signal changing?
Signal is changing how its apps respond to on-device AI on each major platform. According to the post:
- iPhone (iOS 27): Signal prevents screenshots within the app, and users will find a "Block Siri AI" setting.
- Android: When Signal detects an AI assistant trying to read the screen, it turns on "Screen security" automatically. Signal said it cannot reliably detect screenshot attempts on every manufacturer's Android phones, so it chose this route to avoid breaking ordinary screenshots for everyone.
- Mac (macOS): Signal said it cannot currently block screenshots on macOS, so it is not blocking Siri AI there. It recommends that users add Signal to the Excluded Apps list under System Settings > Siri.
Users can switch the protections off, but Signal warned that doing so may expose their own and their contacts' messages to AI agents. The company also acknowledged limits: blocking screenshots does not stop other routes, such as reading content already stored on the device through the filesystem.
Which AI tools is Signal worried about?
The post names Google's Gemini Intelligence on Android, Apple's Onscreen Awareness and Siri AI on iOS and macOS, and Meta's Muse, a third-party agent that Signal says may use accessibility settings to see other apps. Signal said accessibility permissions were never meant to give AI broad access, and called their use by agents "a skeleton key".
Reuters, in a report republished by TimesLIVE, said it had reported earlier in the week that Meta's Muse agent had reached Signal messages on a Mac and on an Android phone using operating-system accessibility and screenshot features granted during setup. Reuters said an Apple update that week closed that pathway and added new permissions and warnings for agents. Meta told Reuters that agents need to see users' screens to work and that Muse reaches other apps only with the user's explicit permission.
Background: Signal and Microsoft Recall
This is not Signal's first move of this kind. The post recalls that last year Signal used screenshot blocking to stop Microsoft's Recall feature from capturing Signal messages by default on Windows 11. Recall takes a picture of the screen every few seconds, reads the text and stores it in a searchable database on the device, which Signal said could keep its messages, including disappearing ones, indefinitely.
What is Signal asking Apple, Google and Microsoft to do?
Signal is asking the three platform companies for two things. First, a simple, well-documented way for developers to opt their apps out of AI data access and processing, covering screen reading, files, notifications and other system routes, without breaking the app. Second, clearer public technical documentation on how their AI agents access and handle data. Signal said confidential computing, where providers process data in protected cloud hardware, is "not a complete solution" because the providers would still decide the scope of any privacy guarantees.
Signal also wants tools to limit how third-party agents use accessibility permissions. It said it was cautiously optimistic that Google and Apple were engaging on the issue and that it hoped the industry would settle on standard approaches.
Why it matters
End-to-end encryption protects a message while it travels between phones, but not once it is displayed on a screen that an AI system can read. Signal's move highlights a growing tension as phone makers build assistants that act across apps: the more an agent can see, the more useful it is, and the more it can expose. For users in India and elsewhere who rely on encrypted messaging, the practical takeaway is that a phone's AI settings now matter as much as an app's own security. Signal framed itself as "the canary in the coal mine" for the wider app ecosystem.
The debate over how much control platform owners should have over apps is also playing out in other areas, including Google's plan to verify Android developers worldwide from 2027, and in growing concern about AI-assisted cyberattacks.




