Skip to content
LATEST
Technology

Signal limits screenshots to curb AI access to messages

Atlas Newsdesk
Signal limits screenshots to curb AI access to messages

Signal is restricting phone screenshots to limit AI access to chats, saying device-level assistants threaten its privacy protections.

The messaging service announced the measures on November 20, describing them as "incomplete" and seeking stronger safeguards from operating system developers. Its concern extends beyond screenshots: Signal wants applications to be able to refuse AI access through notifications, screen-reading tools and files.

The restrictions will be enabled automatically, Signal said. Depending on the device, attempts to capture a screenshot or let an assistant inspect the display will trigger prompts allowing users to turn the protection off.

Encryption meets device permissions

Signal's warning concerns what happens when messages become visible on a device, rather than whether its encryption has been broken. The service says screen-reading AI can copy conversations outside its protected application and send that information to servers used by the assistant.

Owned by a nonprofit foundation, Signal avoids advertising and minimizes information retained on its servers. It uses end-to-end encryption, which prevents the service itself from reading users' conversations; its warning is that a separate application with screen access presents a different exposure.

AI agents are designed to retrieve information across applications and perform tasks for users, including handling email, scheduling appointments and purchasing goods. That functionality depends on access that can overlap with information a messaging application is designed to protect.

In a post, Signal president Meredith Whittaker identified AI embedded within operating systems as a particular concern, naming Apple's Siri AI and Google's Gemini Intelligence. Signal argued that the companies controlling those systems also determine the conditions under which privacy-focused applications operate.

Signal seeks controls from platform owners

Signal asked Apple, Alphabet's Google and Microsoft to give developers practical tools for excluding their applications from AI data processing. It also sought clearer explanations of how assistants work, so developers can determine which information they access and adapt their protections.

The phone measures follow Signal's effort last year to prevent Microsoft's Recall AI from capturing its Windows application. Signal said it cannot consistently provide equivalent protection on Apple's macOS, leaving the effectiveness of its safeguards dependent on the platform.

Meta offered a different emphasis when addressing questions about its Muse agent: assistants need to observe and operate computers to carry out users' requests. The company said Muse reaches other applications only through permissions expressly authorized by users.

Those positions expose a distinction between permission granted to an assistant and a messaging application's ability to exclude its own contents. Signal is seeking an application-level refusal mechanism; Meta's explanation centers on authorization by the person using the device.

Privacy safeguards depend on operating systems

If platform owners provide the controls Signal requests, the service would have more ways to prevent assistants from processing chats, while developers of other sensitive applications could use similar exclusions. For agent providers, those boundaries would limit which information their tools can use when completing tasks across applications.

If screen blocking remains the principal defense, Signal's acknowledgment of incomplete protection leaves other access routes unresolved. The specific open questions are whether operating system providers will offer broader exclusions and whether those controls will cover notifications and files as well as screenshots.

More stories

Latest news