Signal said on 20 November that it will restrict the AI agents built into phone operating systems from capturing Signal chats. On iOS 27 it will block screenshots inside the app. On Android, where it can detect an AI assistant trying to read the screen, it will switch on its existing “Screen security” setting automatically. The post, signed by Meredith Whittaker, names Gemini Intelligence on Android and Onscreen Awareness with Siri AI on Apple devices.
What changes for users
- iOS 27: screenshots are blocked in Signal by default. You can turn the protection off in Signal’s settings, but Signal warns that doing so may expose your and your contacts’ messages to these agents.
- Android: Signal says many vendor builds give it no reliable way to tell when a screenshot was attempted, so instead it watches for an AI assistant accessing the screen and then enables Screen security.
- macOS: not covered. Signal says it cannot reliably block screenshots on macOS because Apple offers no clearly documented API for it. Its only advice for macOS 27 users is to add Signal to the Excluded Apps list in System Settings, under Siri.
This repeats what Signal did on Windows last year, when it used screenshot blocking by default to keep Microsoft Recall from capturing messages.
Why Signal says it has to
Signal’s argument is that end-to-end encryption stops at the screen. Once an OS agent “scans” what is shown, the data sits in “whatever local database or remote server an OS-level AI agent is using at that moment”, outside Signal’s control, and it affects everyone in a conversation, not just the person who turned the agent on. Signal also calls its fix “incomplete”: blocking screenshots does nothing about agents reading app data through the filesystem.
The post makes two asks of Apple, Google and Microsoft. First, a simple, documented way for developers to opt an app out of AI data access, covering screen reading, filesystem access, notifications and other OS frameworks. Second, real documentation of how these agents access and process data. Signal says it spent “considerable developer time” on forensic testing because that documentation does not exist, and that confidential computing does not solve the problem. It also flags third-party agents such as Meta’s Muse, which it says lean on accessibility permissions “meant to empower actual people”. Signal adds that it is “cautiously optimistic” because Google and Apple are talking with it.
For background on the agents in question, see our coverage of Apple’s Siri AI App Intents preview and of how Android isolates some on-device AI in a protected enclave. Meta’s own Muse data choices came up in our piece on Muse and friends’ data consent.
Why this matters
OS-level agents read whatever is on screen, which quietly removes the privacy line that apps like Signal draw with encryption. Signal’s own testing found no documented opt-out on iOS, Android or macOS, so any app handling health, finance or private messages faces the same gap. Until the platforms ship an opt-out API, check what screen protection your platform offers, decide whether to turn it on by default, and tell users what it costs them.

The Campfire
CommentsThanks. Your comment is waiting for the desk. Once it is approved it appears here within about a minute.
Add a comment