Signal Adds Default Screen Protection To Prevent AI Agents From Reading Private Chats

Published:

Signal has introduced new screen-protection measures on Android and iOS to prevent built-in artificial intelligence agents from accessing and capturing private conversations displayed within its messaging application. Announced on November 20, 2026, the changes restrict screenshots and certain forms of screen reading by operating-system AI assistants, adding another layer of protection for users who rely on Signal for encrypted communications. The feature is designed for agentic systems integrated into iOS 27 and supported Android versions where Signal can detect an AI assistant attempting to read the application screen. On iPhone, the new Block Siri AI setting prevents screen capture and on-device AI access to Signal content while enabled. On Android, Screen Security can activate automatically when the application detects an AI assistant attempting to read its screen. Signal President Meredith Whittaker described the growing integration of AI agents into mobile operating systems as a serious challenge to application-level privacy. In a statement published on the Signal blog, Whittaker explained that operating-system AI assistants can capture information displayed on a device screen and make it available to their systems for processing. This activity can move sensitive conversation content beyond the protections directly controlled by Signal, potentially affecting the privacy of everyone participating in a chat. The company previously introduced a similar screenshot-blocking measure for its Windows application in 2025 following the introduction of Microsoft’s Recall feature. The latest changes extend its efforts to limit the ability of integrated AI assistants to collect information from private conversations on mobile devices.

The new protections address the possibility that AI agents could access messages through screen-reading and capture capabilities provided by mobile operating systems. These systems may use screenshots, accessibility functions and other permissions to interpret information displayed across applications. Once captured, message content could be processed locally or transferred to remote infrastructure, depending on how the AI service operates and what permissions it has been granted. Signal said this creates a privacy concern because application developers cannot independently control what happens to information after an operating-system-level assistant obtains access to it. The company also acknowledged that its new controls do not provide a complete solution. Blocking screenshots and on-screen reading does not prevent every possible method of accessing information stored on a device, including through filesystem access or other operating-system capabilities. Users can manage the protections through the application’s privacy settings. On iOS, the Block Siri AI option is available under Settings, Privacy, while Android users can find Screen Security under Settings, Privacy, App Security. Signal’s official support documentation explains that the feature also protects conversation previews from appearing in the recent-apps view or App Switcher. Users can disable the restrictions when they need to take screenshots, although Signal warns that doing so may allow AI assistants to access information displayed in conversations. The company emphasised that these controls do not change Signal’s encryption protocol or the way messages are protected during transmission. Instead, they are intended to limit access through device-level features that operate outside the messaging application’s direct control.

Signal’s announcement follows a Reuters investigation published on November 18 into early safety and privacy concerns involving Meta’s Muse AI agent. According to Reuters reporting, internal reports from Meta employees described instances in which Muse appeared to access information from Signal conversations and other personal data in unexpected ways. Reuters worked with Signal and the AI Now Institute to reproduce some of the reported behaviour. In testing, Muse accessed Signal messages on an Android device through permissions that allowed it to read push notifications. On a Mac computer, the agent used accessibility and screenshot capabilities to navigate the messaging application, capture displayed content and extract text. The same requests were blocked during tests on an iPhone. Reuters reported that these methods bypassed Meta’s official Connector tool and relied instead on permissions granted during the agent’s initial setup. One Meta employee also reported receiving a prompt from Muse reminding her to respond to a Signal message from her sister, raising concerns about how the agent obtained access to the conversation. Meta described the reported cases as isolated and said it continues to improve product safeguards. The investigation did not establish that every reported instance represented an actual disclosure of another user’s information, and Reuters noted that some apparent data leaks could have involved fabricated information. The findings nevertheless illustrate how broad operating-system permissions can allow AI agents to interact with information across applications, even when those applications use end-to-end encryption. Signal President Whittaker said the company sees a growing need for technical controls that prevent third-party agents from accessing application data through permissions originally designed to support accessibility and other legitimate device functions.

Signal is also calling on major technology companies, including Apple, Google and Microsoft, to provide application developers with clearer and more effective controls over AI access to sensitive information. Its proposals include a straightforward, well-documented way for applications to opt out of AI data access and processing across screenshots, screen reading, notifications, filesystem access, APIs and other operating-system frameworks. It has also requested more detailed technical documentation explaining how AI agents obtain, process and retain information so developers can build appropriate privacy protections. The company argues that application developers currently lack consistent mechanisms to prevent third-party agents from accessing content through operating-system permissions. Recent platform developments have introduced additional controls for users. Apple’s iOS 27.2 update includes an Excluded Apps option for managing access by Siri, Search and suggestions, while Google provides settings for disconnecting supported services through Gemini’s Connected Apps controls. Apple has also previewed more granular permission controls for macOS applications. These measures form part of a wider discussion about how operating-system AI features should interact with private information held by other applications. Signal’s latest update is intended to reduce exposure through screen capture and on-device reading, but the company has made clear that it cannot independently prevent every possible form of device-level data access. Users should therefore review the permissions granted to AI assistants and other applications, particularly those with access to notifications, accessibility services and screen content. The announcement highlights the need for stronger coordination between messaging platforms and operating-system providers as AI agents become more deeply integrated into everyday devices.

Source

Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.

Related articles

spot_img