Signal blocks screenshots to keep phone AI assistants out of chats
Signal is blocking screenshots on iOS 27 and some Android phones to keep built-in AI assistants from capturing chats. Macs and other routes are not covered.
By VibeLeaderboard
Nov 21, 2026, 3:36 PM UTC
Signal, the encrypted messaging app, said on November 20 that it is restricting the AI assistants built into phone operating systems from capturing its users' chats. In a post by Signal's president, Meredith Whittaker, the nonprofit said it is preventing screenshots in its app on Apple's iOS 27, and that on Android it automatically switches on its existing "Screen security" setting when it detects an AI assistant trying to capture the screen. The post names Apple's Siri AI, with a feature Apple calls Onscreen Awareness, and Google's Gemini. Anyone who wants screenshots back can turn the protection off in Signal's settings. Signal took a similar step on Windows last year, blocking screenshots so Microsoft's Recall feature could not record Signal messages by default.1
The move targets the point where encryption stops. The Electronic Frontier Foundation (EFF), a digital rights group, explains that end-to-end encryption helps provide mathematical guarantees that the company running a messaging service cannot read the contents of your messages. But there is no way to guarantee what happens once a message arrives on a phone and is shown on its screen.2 Signal's post is about assistants that work at exactly that point.1
Both companies describe assistants that use what is on your screen. Apple says Onscreen Awareness lets Siri AI "answer questions or take actions related to the content on a user's screen," such as finding the home games listed on a sports team's website and adding them to a calendar.3 Google's help page says that when you tap one of Gemini's on-screen suggestions, Gemini "will upload the info on your screen" to use as context, and that with a setting called Intelligent Screen Context turned on, Gemini "can infer when it should use" what is visible on the screen.4 Signal says that when these assistants "scan" the screen inside its app, the messages are captured and made available to the assistant so it can answer questions and carry out tasks.1
Apple's announcement does not settle what happens to that captured text afterward. Apple says its new models run both on the device and on its own servers through a system called Private Cloud Compute, and that some Siri AI components, such as its search index, work entirely on the device. It does not say which Siri AI or Onscreen Awareness requests stay on the phone and which go to its servers. Its general promise is that when Private Cloud Compute handles a request, personal data "is not stored nor made accessible to Apple or anyone else."3 EFF, writing about such server systems in general, calls them "privacy-preserving" but not "encrypted," and argues that when a phone sends messages to one automatically, the system as a whole is no longer end-to-end encrypted.2 Signal's own claim is broader: it says message data is being "captured, stored, and made accessible to whatever local database or remote server an OS-level AI agent is using at that moment," beyond its control.1 These are separate claims, not an exchange: Apple and Google did not immediately respond to Gizmodo's requests for comment on Signal's move.5
Signal's post does not describe any case in which an assistant exposed someone's messages. Its argument concerns access and control: once the operating system has captured the screen, Signal can no longer promise what happens to the text. Signal also says it has spent months on what it calls forensic investigations of Android and iOS to learn what these assistants actually do with app data, citing what it calls a significant lack of documentation from the companies.1
The block is aimed at the route Signal says is used most: screenshots, which it describes as how most of these AI systems get data out of apps like it.1 The Android version works differently because of a limit Signal describes on many manufacturers' phones: once screenshots are already disabled, Signal has no reliable way to tell that someone tried to take one. Blocking screenshots all the time would leave people confused when screenshots simply stopped working, so Signal instead watches for an assistant reaching for the screen and turns Screen security on when it sees one.1 Signal acknowledges, and Gizmodo's report repeats, that the protection will not work on every Android device and that Signal has no reliable way to block screenshots on Mac computers.15
Some routes are outside what the block addresses. Signal calls its remedy "incomplete" because blocking screenshots does nothing about an assistant reading content already stored on the device's file system.1 Notifications are another. In a guide published on October 28, before Signal's change, EFF reported that Gemini cannot access Signal's content directly, because Signal does not support Gemini's app connections, but that it can read incoming notifications if a feature called Device Assistance is switched on. EFF also said Gemini can use screen context unless that is disabled.6 Signal's Android change now targets that screen route on phones where it can detect the assistant. Google's settings list screen text and screenshots as separate switches, and Signal's post does not say whether its block also covers an assistant reading on-screen text.41 Signal adds that in many cases third-party agents, naming Meta's Muse, reach app data through accessibility settings that users are encouraged to grant during setup, settings that Signal says were meant to help people get the most out of their apps.1
The protection also runs separately on each person's phone, which matters for anyone in a group chat. Signal warns that switching the block off can expose "your (and your contacts')" messages to these assistants, and that screen capture endangers "all the participants in a conversation."1 If one friend in a group turns screenshots back on, reads the chat on a Mac, or uses an Android phone where Signal cannot detect the assistant, the messages you sent can be exposed to that route on that device, whatever you have set on your own.
Signal's larger complaint is that apps have no proper way to opt out. It says Apple, Google and Microsoft give developers no easy means of keeping AI assistants away from app data without "drastic compromises" to how an app works, which is also why it cannot yet protect Mac users. It asks for a documented switch that would cover screen reading, file access, notifications and other system features, plus clear technical descriptions of how the assistants handle data. Without naming any company's system, it says "confidential computing" approaches are not a complete solution, because they still leave the providers to decide how much privacy they guarantee. Signal also says it is "cautiously optimistic" because Google and Apple are discussing the problem with it.1
Individual users already have some controls. Apple's iOS 27.2 update added a setting, under Settings > Siri > Excluded Apps, that EFF says stops Siri AI from accessing an excluded app's contents; Onscreen Awareness no longer works there, which EFF reads as a sign the setting is effective.6 Because EFF describes that exclusion as covering an app's contents, adding Signal to the list looks like a useful second layer alongside Signal's own screenshot block, though no source tests the two together. Signal recommends the same Excluded Apps setting, in System Settings > Siri, to people using macOS 27 on a Mac, where it has no block of its own.1 On Android, EFF says Gemini offers no per-app control over screen context. Its guide points to Device controls > Screen context > "Use screen and app data" in Gemini's settings, and to Settings > Personal Intelligence > Connected Apps to check whether Device Assistance can read notifications.6 Google's help page describes the Intelligent Screen Context controls as two switches, "Use text from screen" and "Use screenshot," so the wording in your app may not match both descriptions.4
End-to-end encryption still protects a Signal message on its way between phones. The exposure begins after the message is decrypted and displayed, on the device, through screen capture, stored files, notifications and permissions that Apple and Google control and that Signal says they have not fully documented. Signal's block is designed to close the route it says assistants use most, on iOS 27 and on Android phones where it can spot the assistant. Unless the operating system makers give apps a real opt-out, the privacy of a conversation also depends on the devices and settings of everyone in it.
Sources (6)
Signal, "Signal is Serious About Protecting Your Privacy," Meredith Whittaker, November 20, 2026. Open source
Back to text ↑Electronic Frontier Foundation, "Secure Messaging and AI Remain in Conflict, Despite the Promise of TEEs," Erica Portnoy and Thorin Klosowski, September 18, 2026. Open source
Back to text ↑Apple Newsroom, "Siri AI, a profoundly more capable and personal assistant powered by the next generation of Apple Intelligence, is here," September 14, 2026. Open source
Back to text ↑Google Gemini Help, "Use screen actions to chat with the Gemini mobile app." Open source
Back to text ↑Gizmodo via AOL, "Signal Is Fighting to Keep AI Agents Out of Your Private Messages," Bruce Gil, November 20, 2026. Open source
Back to text ↑Electronic Frontier Foundation, "Victory! You Can Now Exclude Apps From Siri AI," Thorin Klosowski, October 28, 2026. Open source
Back to text ↑
Related Intel (5)
- Signal Is Fighting to Keep AI Agents Out of Your Private Messages
- Use screen actions to chat with the Gemini mobile app
- Victory! You Can Now Exclude Apps From Siri AI
- Secure Messaging and AI Remain in Conflict, Despite the Promise of TEEs
- Siri AI, a profoundly more capable and personal assistant powered by the next generation of Apple Intelligence, is here
Checking sign-in…
Loading comments…