Signal now blocks screenshots by default inside its app on iOS 27, and switches on its “Screen security” mode on Android when it detects an AI assistant trying to read the screen. In a blog post on November 20, Signal president Meredith Whittaker said AI agents built into phone operating systems are an “existential threat” to the app’s privacy promise, and admitted the new protection is only a partial fix.
The move puts a hard question in front of everyone building with agents: if an assistant can read any screen, who decides what it is allowed to see?
Key takeaways
- On iOS 27, a new “Block Siri AI” setting is on by default and stops screenshots inside Signal.
- On Android, Signal turns on its existing “Screen security” setting when it detects an AI assistant reading the screen.
- Users can switch the protection off, but Signal warns this can expose their contacts’ messages too.
- There is no protection on macOS yet, because Signal says it cannot reliably block screenshots there.
- Signal wants Apple, Google and Microsoft to give every app a documented way to opt out of AI data access.
What happened
Apple and Google are building assistants that look at whatever is on your display so they can answer questions and take actions for you. Signal names two of them: Onscreen Awareness in Apple’s Siri AI and Gemini Intelligence on Android. To do their job, these systems capture what is on screen, and that capture can include an open Signal conversation.
Signal’s answer is to treat AI screen capture the way it treated Microsoft Recall last year. In 2025 the Windows 11 app started using a DRM-style flag so that Recall, which snapshots the screen every few seconds, only saw a blank window. Now the same idea is coming to phones.
The details differ by platform. On iPhones running iOS 27, a new “Block Siri AI” toggle is enabled by default and screenshots inside the app are blocked. Android is messier: many device makers ship their own versions of the OS, and Signal says it cannot reliably tell on all of them when a screenshot was attempted. So instead of silently breaking screenshots, Signal watches for an AI assistant trying to access the screen and then turns on Screen security automatically. On both platforms, users see a prompt offering to turn the protection off, as Engadget reported.
Signal also called out third-party agents. It pointed to Meta’s Muse, which Reuters found could read Signal messages through permissions granted during setup: notification access on Android and accessibility permissions on Mac. Signal argues accessibility features were built to help people use their devices, not to act as “a skeleton key” for AI agents.
Why it matters
End-to-end encryption protects a message while it travels. It does nothing once the message is decrypted and drawn on your screen. If an operating system agent copies that screen and sends it to a model, the message now sits in a local database or on a remote server that neither you nor Signal controls. Signal stresses that this affects everyone in the chat, not only the person who turned the assistant on.
The company is open about the limits of its fix. Blocking screenshots does not stop an agent from reading notifications or pulling data already stored on the device through the filesystem. It also does not help if the other person in the conversation has switched the protection off. And on macOS, Signal says it has no reliable tool at all: the screen-protection options Apple offers are aimed at streaming video and DRM content, not at stopping screenshots of an app. Its only advice for Mac users is to add Signal to the Excluded Apps list under System Settings, then Siri.
Signal also says it spent months on forensic work across Android and iOS just to learn what these assistants actually read. That points to the bigger complaint: platforms are shipping agent features without clear documentation of what data they touch.
What it means for developers and businesses
This is not only a messaging-app problem. Any app that shows sensitive data, such as banking, health, legal, HR or internal dashboards, is in the same spot. Today there is no simple, documented switch that says “AI agents may not read this app” across Android, iOS, macOS and Windows. Developers are left with blunt tools like secure-window flags, which also break screenshots, screen sharing or accessibility for real users.
Signal is asking the OS makers for two things:
- An easy, documented way for apps to opt out of AI data access, covering screen reading, files, notifications, APIs and other system frameworks, without crippling the app.
- Clear technical documentation of how these agents access and process data. Signal adds that confidential computing alone is not enough, because the platform still decides how private the data really is.
If you build apps with sensitive screens, a few practical steps make sense now:
- Audit which screens show data you would not want copied into an assistant, and use the platform’s secure-window options there where the trade-off is acceptable.
- Keep sensitive content out of notification previews, since notification access is one of the routes agents use.
- Track OS-level exclusion controls, like the per-app Siri exclusion Apple already offers, and explain them to your users.
If you build agents, the lesson runs the other way. Agents that work by reading screens, like the Gemini computer use feature that just reached the Mac, will increasingly meet apps that refuse to be read. Designing agents to respect app-level opt-outs, and to use proper APIs where they exist, matters more as agents start handling actions and payments, as with ChatGPT Wallet.
What to watch next
Signal says it is “cautiously optimistic” because Google and Apple are engaging with it on how their systems should work with privacy-focused apps, and it hopes to share more soon. Neither company had announced a new developer control at the time of writing. Fast Company and Reuters both framed the update as a sign of growing tension between app makers and platform AI.
Three things are worth tracking. First, whether Apple adds a real screenshot-protection API on macOS, which Signal says it needs before it can protect Mac users. Second, whether Android gets a standard way to detect and block assistant screen reading across all device makers. Third, whether other privacy-sensitive apps follow with their own default-on blocks, which would push this from one app’s setting toward an industry standard.
For users, the short version is simple. On iOS 27 and Android, Signal’s new block is on by default, and leaving it on keeps your conversations, and your contacts’ messages, out of the assistant’s view. On a Mac, the Siri exclusion setting is the only option for now.