SAN FRANCISCO – Signal is drawing a hard line against ambient artificial intelligence, rolling out new security updates to prevent virtual assistants from sweeping up encrypted chats on mobile phones.
The nonprofit messenger announced Friday that it will block screenshots on iPhones running iOS 27 and restrict automated screen capture on select Android devices. The defense mechanism addresses a loophole where AI tools harvest text directly from phone notifications or accessibility APIs.
Closing the Notification Loophole
End-to-end encryption secures messages while in transit, but once delivered, the receiving device decrypts the text for the user. Virtual assistants integrated into operating systems frequently leverage broad permissions to monitor on-screen content and notifications for contextual prompts.
Signal President Meredith Whittaker highlighted the inherent danger of granting background utilities unrestricted viewports. Accessibility permissions originally designed for users with disabilities are functioning as a master key for automated data collection agents.
Privacy is our mission; it’s why we exist, and we do everything we can to protect it, wrote Signal President Meredith Whittaker in a blog post outlining the changes.
On Apple devices operating on iOS 27, Signal’s new Block Siri AI protocol is active by default. Similarly, supported Android hardware automatically engages screen protection routines upon detecting an assistant attempting to parse chat windows.
Industry Pressure on Big Tech
The privacy crackdown follows rising scrutiny over how major tech firms handle third-party application data. Recent internal reports revealed instances where corporate AI tools parsed private messaging alerts without explicit real-time user prompts, relying instead on broad setup permissions.
Signal leadership is now pressing Apple, Google, and Microsoft to establish explicit boundaries regarding what ambient assistants are permitted to ingest. Developers require granular controls to prevent automated screen-scraping from undermining zero-knowledge messaging architectures.
Users retain the ability to toggle off these protections within the application settings menu, though doing so restores an assistant’s capability to read both sides of an active conversation.