The encrypted messaging service is turning off screen captures by default to stop phone assistants from sweeping up private conversations.
Signal has updated its mobile apps on iOS and Android to stop artificial intelligence assistants from reading user conversations without direct consent.
As reported by The Register, the encrypted messaging service introduced default screenshot blocking to prevent tools like Apple’s Siri and Google’s Gemini from scanning chat windows and sending private data to outside servers.
Signal considers automated phone assistants an existential threat to messaging security, Engadget reported. The changes follow rising concern over operating system tools designed to monitor active screens.
Why AI agents pose privacy risks
Phone makers are embedding agentic features directly into their operating systems, including Google’s Gemini Intelligence and Apple’s Onscreen Awareness for Siri AI. To answer questions and assist users, these tools routinely scan whatever appears on the screen.
The assistants extract text from captured displays using optical character recognition. That process means conversations meant to stay encrypted can be collected and moved to corporate servers outside Signal’s control.
Third-party agents pose similar issues across devices. Meta’s Muse was previously caught reading Signal conversations through setup permissions, accessing notifications on Android and tapping accessibility features on Mac.
How the defenses work and where they fall short
The updated protections function differently depending on the operating system. On iPhone, Signal turned on a new Block Siri AI setting by default. On Android, the app relies on its existing Screen security setting to block assistants when they attempt to capture the screen.
Users see a prompt to switch off the guard if the app detects a screenshot on iOS or an assistant scanning the display on Android. People can disable the restriction in Signal’s settings menu, though Signal warned that doing so risks exposing their messages and their contacts’ chats to AI agents.
The defense has clear limits. If a contact disables the safeguard on their own phone, their device can still let an AI assistant read the conversation. The setting also cannot stop software from inspecting device files or reading notifications.
Signal CEO Meredith Whittaker warned that developers have few options, saying, “it is a serious problem that Signal – and application developers beyond – lack robust tools to prevent AI agents from undermining application-level privacy.”
Whittaker noted that operating system creators do not provide easy ways to restrict AI data collection without hurting the core user experience, which is why Signal cannot offer the protections on macOS.
A growing backlash against agent software
Signal previously took steps against system-wide screen capture last year, when it blocked Microsoft’s Recall tool from taking snapshots of conversations on Windows 11.
The pushback against automated agents extends well beyond private messaging. Multiple companies have introduced blocks to keep bots and agents from reading their data or interacting with their platforms. Amazon barred shopping agents like Meta Muse from its retail site while permitting them on AWS Marketplace, and United Airlines prohibited transactions from unauthorized bots.
Other businesses establishing barriers against third-party AI agents include Target, Home Depot, Etsy, eBay, JPMorgan Chase, Charles Schwab, Citigroup, and Wells Fargo. The restrictions have fueled business for bot detection and mitigation firms like Cloudflare, Akamai, and DataDome.
What comes next for operating systems
Apple is testing macOS 27.3 in public beta, which adds more detailed controls for its Device Control and Data Access API and its Full Disk Access API. Those changes will trigger more permission warnings for Mac programs and require backup tools and accessibility software to get special entitlements from Apple.
Introducing the disk access change, Apple stated, “As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially.”
Signal wants operating system vendors to establish a standardized, system-level method that lets developers stop AI agents from reading app data. Signal said it is currently talking with Apple and Google about the issue and feels cautiously optimistic about the discussions.
