Signal has cut off AI agents' access to the screen in its mobile apps. We're talking iOS and Android: a new toggle in the settings stops the system from capturing and parsing what's happening in your chats. The company explained the move in a blog post, which GIGAZINE summarized on November 22, 2026.
This is round two, not round one. Back in May 2025, Signal on Windows already blocked screenshots over the Recall feature in Windows 11, which captures the screen and parses your activity history. After security complaints, Microsoft disabled Recall by default. At the time, Signal said Microsoft had left it no choice — and now the same logic has moved to phones.
Here's the mechanics: an agent reads the screen, extracts text and images, and either stores them in a local database or ships them to a server. It then answers questions and executes commands — but outside Signal's encryption. The blog calls this an "existential threat": Google is building in Gemini Intelligence, Apple is building in Siri AI with Onscreen Awareness, and both reach exactly the layer where third-party apps live. In Signal's framing, these operating systems are the sea the messenger swims in — and the owner of the sea sets the rules of survival.
Hence two demands for Apple, Google, and Microsoft. First: a simple, documented way to exclude an app from screen reading, file access, notifications, and APIs — without gutting the app's usefulness. Second: clear technical documentation of agent behavior, especially data access. Signal dismisses confidential-computing approaches out of hand: the right to decide where the privacy boundaries lie stays with the provider.
This news didn't land in a vacuum. Agents have stopped being a feature of a single app and become an OS layer — which means an app no longer controls where what the user sees ends up. Signal is pulling the only lever it has: its own window. Our read: this isn't a technical solution, it's a statement of position.
What changes for you and me. App privacy rested on a contract: "the developer is responsible for data inside their own process." Agents break that contract — they look across processes and stash what they see wherever the developer can't reach. Signal's toggle is, in essence, an attempt to restore the old model by hand, and it only works as long as the user doesn't switch it off themselves.
Who wins? Signal users who care about their conversations — they get at least some protection. Who loses: third-party apps in general. The deeper agents embed into the OS, the less their own guarantees are worth. We think the real fight won't be over screenshots — it'll be over who gets to set the "this app is off-limits to agents" flag. As long as that flag is set by the developer rather than the OS, there's no real protection.
The trigger contains no numbers, no iOS or Android versions, no technical details: how exactly the OS blocks screen capture, or what happens to files already saved. Signal says it ran forensics on both platforms but isn't publishing the results. There's also no response from Apple, Google, or Microsoft to the two demands. All of this is one blog post and its retelling — not confirmed cases of leaks via an agent. Worth checking yourself: is the toggle on, and what's actually asking for screen access.
Forecast: within a quarter, Apple or Google will respond to Signal's request — at least with a line in developer documentation. My money's on documentation, not a new API: it's cheaper and commits to nothing.
Disclaimer: Not officially confirmed: everything rests on a single Signal blog post and GIGAZINE's retelling — no iOS or Android versions, no technical details of the blocking, no forensic results, and no response from Apple, Google, or Microsoft.
Source: GIGAZINE
Source: gigazine.net · post in Telegram