Signal blocks phone AI agents from reading chats, calling them an “existential threat” to privacy

Encrypted messaging app Signal announced on November 20 that it will block built-in AI assistants in phone operating systems from reading users' chats by default. On iOS 27 it blocks in-app screenshots and adds a “Block Siri AI” toggle; on Android it turns on screen security automatically when an AI assistant tries to read the screen. Signal president Meredith Whittaker called such “agentic AI integration” an “existential threat” to privacy.
Encrypted messaging app Signal announced on November 20 that it will block AI assistants built into phone operating systems from reading users' chats by default. Apple, Google and Microsoft are weaving “agentic AI” deep into their operating systems, Signal said, and those systems scan what is on screen and store it — undermining privacy guarantees at the application layer. Signal president Meredith Whittaker called the integration an “existential threat” to privacy.
Screenshots blocked on iOS 27, screen security on Android
On iOS 27, Signal is preventing screenshots inside the app and adding a “Block Siri AI” setting. On Android, the app automatically turns on its “Screen security” protection when it detects an AI assistant attempting to read the Signal screen. People can switch the protections off in settings, Signal said, but doing so could expose their own and their contacts' messages to these AI agents and break its privacy guarantees.
Signal already moved against Microsoft Recall
This is not the first time Signal has taken such a step. Last year it used screenshot blocking to stop Microsoft's Recall AI from capturing and storing screenshots of Signal messages on Windows 11 by default. The problem is the same, Signal said: operating-system AI agents “scan” the information on screen, capture it and make it available to agentic systems so they can answer questions and carry out instructions — and once captured, the data is outside Signal's control, stored in whatever local database or remote server the agent uses.
The fix is incomplete, and macOS is out of reach for now
Signal acknowledged the remedy is incomplete. Blocking screenshots does not stop AI agents from reaching app data through other, sneakier routes, such as reading content already on a device via the filesystem — and the company said it cannot currently implement these protections on macOS because the system provides no way to do so. Over the past few months Signal said it spent considerable developer time on forensic investigations across Android and iOS to learn what actually happens when these AI systems access app data, because documentation is badly lacking.
A call for opt-out tools from the big three OS makers
Signal called on Apple, Google and Microsoft to provide two “common-sense remedies”: an easy, well-documented way for developers to opt their app out of AI data access and processing — including screen reading, filesystem access, notifications, APIs and other OS frameworks — without undermining the app's core utility; and better documentation and technical specifications on how these AI agents work and handle data. Confidential computing is not a complete solution, Signal argued, because it still leaves the power to determine the extent of privacy guarantees in the providers' hands.
Not just Signal: third-party AI agents are expanding too
Signal also flagged a growing need for tools to stop third-party agents, such as Meta's Muse, from accessing app data using the deep permissions users are encouraged to grant during setup — in many cases leveraging accessibility settings meant to empower people with disabilities. Signal thanked allies including the Electronic Frontier Foundation for pushing similar changes, and said it is the “canary in the coal mine”: integrating AI agents into the operating system without common-sense remedies is an existential threat to privacy across the whole app ecosystem.
N環球 
💬 Comments
No comments yet. Be the first!