Signal President Meredith Whittaker: System-level AI Poses a Survival Threat to Privacy
Signal announced on its official blog today that, in order to prevent the operating system’s built-in AI agents from capturing and storing users’ conversations, it will by default restrict screenshots. On iOS 27, screenshots will be directly blocked, including in the App Switcher; on Android versions that can detect AI assistants reading the screen, “Screen security” will be automatically enabled. Users can turn it off in the app settings, but doing so will expose their own and their contacts’ content to AI agents and weaken privacy protections.
Meredith Whittaker made it clear that once capabilities such as Google Gemini Intelligence and Apple Siri AI’s Onscreen Awareness are deeply embedded in the operating system, agents will scan screen information, causing conversations to enter local databases or remote servers. This process goes beyond the controllable scope of Signal’s open-source code and encryption protocols. Last year, Signal had already taken similar screenshot-blocking measures against Microsoft Recall on Windows. The protection is not complete and cannot cover ways such as reading existing content from the file system, notifications, and accessibility permissions; at present, screenshots also cannot be reliably blocked on macOS, and macOS 27 users are advised to add Signal to Siri’s “Excluded Apps.”
Whittaker described this as a “survival threat to Signal and its mission,” and called on Apple, Google, and Microsoft to provide clear developer opt-out mechanisms and technical documentation, while also preventing third-party agents such as Meta’s Muse from reading app data through accessibility permissions. Signal has already been in discussions with some system vendors, but emphasized that there are currently no standardized tools.
Source: Public information
ABAB AI Insight
The operating system is shifting from an “application runtime container” to a “continuous sensing and action layer.” A historical precedent is Microsoft Recall’s continuous screen capture on Windows 11, and Signal had already responded at the time with screenshot blocking. Extending the same logic to iOS 27 and detectable Android environments shows that end-to-end encryption has already failed at the layer where “screen-visible content is captured by system-level agents.” In terms of personnel, Whittaker is continuing her long-standing criticism of system vendors’ data control, forcing privacy protection upward from the application layer to the operating system layer.
On the capital side, Google, Apple, and Microsoft are embedding agents deeply into the OS, re-centralizing data control rights that originally belonged to application developers. Signal, as a non-profit privacy tool, is forced to defend itself at the cost of user experience (blocking screenshots by default), while system vendors can continue collecting contextual data through agents for model improvement and service loops. Third-party agents such as Meta’s Muse, meanwhile, bypass application boundaries by leveraging accessibility permissions, creating new data extraction channels.
Compared with last year’s Windows case, this time the coverage spans both major mobile platforms and is enabled by default, expanding the defensive scope, but Whittaker explicitly acknowledges that it remains incomplete. Conclusion: application-layer privacy boundaries are being systematically eroded by operating system agents; if system vendors do not provide verifiable opt-out interfaces and documentation, independent privacy apps will continue to be in a passive defensive position, and data control will further concentrate at the platform level.
Cognitive laws:
What is visible on the screen can be captured by the system
Encryption protocols end at the operating system boundary
The right to privacy protection shifts from applications to platforms