On November 20, 2026, Meredith Whittaker, president of the encrypted messaging app Signal, announced on the official blog that Signal is starting to block the AI agents built into phone operating systems from reading users' chats by default. On iOS 27, screenshots inside the app are blocked by default, with a new "Block Siri AI" toggle; on Android, Signal automatically turns on screen security when it detects an AI assistant trying to read the screen. Shut out are Apple's Siri AI onscreen awareness and Google's Gemini Intelligence — both system-level assistants "look" at the screen to help you. Signal believes this is making the privacy environment increasingly hostile, even threatening its very reason to exist as a privacy app.
What exactly Signal changed
This is not the first time Signal has done this. Last year, it used screenshot blocking to stop Microsoft's Recall AI on Windows 11 from storing chats in a system database. The approach this time is similar, only the battleground has moved to phones: on iPhone, screenshots are blocked by default plus an explicit toggle; on Android, screen reading by an AI assistant triggers automatic hardening. Users can also turn this protection off manually in settings, but Signal warns clearly: once off, your chats — and your contacts' — can be captured by AI assistants. On macOS this approach is not possible for now, and Signal advises users to exclude Signal from Siri in the system settings themselves.
What this defense can — and cannot — stop
Signal puts it bluntly itself: screenshot blocking closes only one route. AI assistants can still reach content already on the device through the filesystem, and they can read notifications; third-party agents are worse — a Reuters report this week noted that Meta's Muse reads Signal messages using permissions granted during setup, reading notifications on Android and using accessibility permissions to operate other apps on Mac. If the other person turns the protection off, your defense is useless too. Signal is therefore asking Apple, Google and Microsoft for two things: a simple, documented way for developers to let an app opt out of AI data access — including screen reading, filesystem access, notifications and other interfaces — and clear technical details on how AI assistants access and process data.
For ordinary users, this update requires almost no action: update Signal to the latest version and the protection is on by default. But for people with higher privacy needs, such as journalists and activists, Signal's warning deserves a closer look — a screenshot ban stops system assistants from reading the screen, but not the settings on the other person's device, and not third-party agents that already hold high privileges. What would really fix this is an opt-out mechanism from the operating system makers, and that is exactly why Signal is speaking out publicly. System-level agents are also moving from "reading the screen to help" toward "operating the computer directly": Apple's own computer-use agent is unlikely to arrive before late 2027, while Gemini computer use has already landed on Mac — check region and subscription before turning it on.