Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Signal blocks screenshots by default to keep OS-level AI agents out of chats

On November 20 Signal said it will block in-app screenshots by default on iOS 27 and enable screen protection on Android when an AI assistant tries to read the app, arguing that OS-level agents capture chat content and hand it to local or remote databases. Signal concedes the remedy is incomplete and unimplementable on macOS, and calls on Apple, Google and Microsoft to give developers tools to opt out of AI data access.

Published
Signal 默认拦截截图,阻止系统级 AI 智能体读取聊天
Image source: signal.org

On November 20, Signal president Meredith Whittaker announced in a company blog post that the messaging app will restrict operating-system AI agents from capturing user data, blocking in-app screenshots by default on iOS 27 and switching on protection on Android when an AI assistant attempts to read the screen.

Signal has done this before. When Microsoft shipped Recall on Windows 11, Signal used screenshot blocking to stop the feature from recording Signal chats by default. This time the target is Apple and Google, whose iOS 27 Siri AI offers Onscreen Awareness and whose Gemini Intelligence is woven deeper into Android.

The mechanism is straightforward. Those system-level agents scan what is on the screen and hand captured Signal content to a local database or a remote server. Once captured, the data sits outside Signal's control, which is precisely the gap that end-to-end encryption cannot close.

So Signal is blocking collection at the source. iOS 27 gets a hard block on in-app screenshots. On Android, where vendor variants make screenshot attempts hard to detect reliably, Signal instead enables its Screen security setting automatically when it detects an AI assistant reaching for the screen. Users can turn the protection off in settings, but that exposes their own messages and those of their contacts.

Signal is candid that the remedy is partial: screenshot blocking does not cover filesystem reads or other quieter routes, and macOS cannot be covered at all because Apple offers no usable API for blocking screenshots. The only advice for Mac users is to add Signal to the Excluded Apps list under Siri in System Settings.

The deeper complaint is about tooling. Whittaker argues that Google, Apple and Microsoft give developers no way to limit AI data access without wrecking the core user experience, and that documentation on how these agents work is thin. Signal says it has spent months of developer time doing forensic investigation across Android and iOS just to learn what happens when the agents touch app data.

Signal's asks are specific. Developers should get an easy, well-documented way to opt an app out of AI data access and processing, covering screen reading, filesystem access, notifications and APIs. Operating-system vendors should publish technical specifications for how their agents access and process data. The post also flags third-party agents such as Meta's Muse leaning on accessibility permissions that were meant for people, not for automation.

For users, the immediate change is that screenshots stop working on iPhone until the setting is flipped. For the industry, the post frames the stakes bluntly: encryption protects the wire, but privacy at the application layer now depends on whether platforms hand developers a switch. Signal says talks with Apple and Google are ongoing, and whether a standardized opt-out emerges is the thing to watch.

Why it matters

Encryption protects the wire, not what an on-screen agent reads; Signal's move turns the permissions question around operating-system AI from a debate into a concrete product conflict.

Signal隐私AI Agent
Back to realtime news

Nearby Updates

All

11/21, 00:41

Apple reportedly building an AI agent that can write code and operate apps

Reports on November 20 said Apple is developing an AI agent capable of writing code and using apps to carry out tasks, putting it in the same category as agent products from rival AI companies. The coverage was also read as a signal that Apple does not intend to sit out the AI race, though no timeline, model source or product form has been disclosed.

11/21, 01:49

Arena Revises Alignment Index Method, OpenAI Takes All Seven Top Spots

Arena updated the methodology behind its Alignment Index on November 20, saying the revised signals catch model failures more broadly and reliably, with OpenAI models holding all seven top positions. The index scores models on unauthorized actions, false attribution and deceptive completion seen in real agent sessions, making such rankings highly sensitive to how failures are defined.

11/21, 01:53

Google's Gemini clears the static from firefighter radio, cutting Watch Duty alerts to under 30 seconds

Google detailed on November 20 how a Gemini-powered tool built with the nonprofit wildfire app Watch Duty transcribes noisy fire radio into clean, searchable text and flags new dispatches in under 30 seconds. Word error rate on degraded VHF audio fell from 38% to 22.34%, and volunteer reporter satisfaction with transcription rose from 35.3% to 92.3%.

11/21, 02:15

METR's updated time horizon plot: exponential gains continue, judgment lags

METR has published an updated version of its widely cited and widely misread time horizon plot, adding longer and more open-ended tasks to probe how models handle work that cannot be scored automatically. The report finds capabilities on verifiable tasks still improving exponentially, while models remain below human level on judgment-heavy work such as designing evaluations or scoring philosophical arguments.