Saturday, Nov 21 | --:--
Back to home

Signal Blocks Screenshots by Default to Keep OS AI Agents Out of Chats, Admits Gaps

Signal said on November 20 that it is blocking screenshots by default in its app on iOS 27, and switching on its Screen Security setting on Android when it detects an AI assistant trying to read the screen, to stop agents such as Siri AI and Gemini from capturing chats. Users can turn the protection off. Signal calls the remedy incomplete, says it cannot offer it on macOS, and asks Apple, Google and Microsoft for a documented way for apps to opt out of AI data access.

4 min read New York View as Markdown
A handmade editorial collage on torn ivory paper in cobalt and vermilion: a big cobalt cut-paper roller blind hangs from a slim rolled tube in front of a tall pile of blank folded paper letters, with the bottom of the pile peeking out under its ragged hem, while a long flat cut-paper arm and open hand reaches in from the left and presses against the blind, stopped; an official Signal logo is glued on a torn-paper clipping whose right end is tucked under the blind's left edge, the full logo visible; tiny vermilion scraps in the corners.

End-to-end encryption protects a message on its way between phones. It does nothing once an AI agent built into the phone reads the screen, and Signal has now said so in product, not just in speeches. The encrypted messenger said on November 20 that it is blocking screenshots by default in its app on iOS 27, and on Android where it can detect an AI assistant trying to read the screen, to keep operating-system agents such as Apple's Siri AI and Google's Gemini out of users' chats.

The more telling part is what Signal admits it cannot do. It calls the remedy "incomplete," says it cannot offer it on macOS at all, and argues that only Apple, Google and Microsoft can fix the underlying problem.

What Signal changed#

In a post by president Meredith Whittaker, Signal says it is "preventing screenshots in the app on iOS 27 and when AI assistants attempt to capture screen data on Android." On Android, where many device makers give it no reliable way to detect a screenshot, it instead turns on its existing Screen security setting automatically when it detects an AI assistant accessing the screen. Reuters, which saw the post before release, reported that the restrictions are on by default and that users get a pop-up offering to disable them.

People who want screenshots back can switch the protection off in Signal's settings, but Signal warns that doing so "will potentially expose your (and your contacts') Signal content to these AI agents, and break our privacy guarantees." The company did the same thing on Windows last year, when it used screenshot blocking to keep Microsoft's Recall from capturing Signal messages.

What it cannot block#

Signal is candid about the gaps. Blocking screenshots "does not address other more sneaky ways AI agents could potentially access application data, such as retrieving content already present on a user's device via the filesystem," the post says. It cannot cover every Android version, and it is "currently unable to implement these protections on macOS" because Apple offers no documented way to stop screenshots there. For Mac users it suggests adding Signal to the Excluded Apps list in the Siri settings.

That is why the post ends in a request rather than a fix. Signal asks the three operating-system makers for "an easy and well-documented way to opt their app out of AI data access and processing," covering screen reading, filesystem access, notifications and APIs, and for documentation of how their agents actually handle app data. It says some providers, naming Google and Apple, are engaging with it. Apple's own agent push makes the request concrete: earlier this month it opened Siri AI to every developer so the assistant can run actions inside other apps.

The Muse precedent#

Signal's post also names a third-party agent. It says tools are needed to stop agents "like Meta's Muse" from using the deep permissions users grant at setup, including accessibility settings, as "a skeleton key." That follows a November 18 Reuters investigation in which testers working with Signal staff and the AI Now Institute got Muse to read Signal messages on an Android phone, through its notification permission, and on a Mac, through accessibility and screenshot tools, bypassing Meta's own "Connector" approval route; the same requests were blocked on an iPhone. Meta told Reuters that Muse can access other apps only through permissions users explicitly grant. Reuters also reported that an Apple operating-system update this week closed the specific Mac pathway and added a new set of agent permissions with extra warnings.

Our read: Signal is now treating the phone's own assistant as something to defend against, and its admission that it cannot finish the job on its own turns a privacy setting into a platform-policy question for Apple and Google. Whether they ship an app-level AI opt-out is the thing to watch.

Limits#

  • Signal's description of how operating-system agents capture and store screen data is its own; Apple, Google and Microsoft had not responded publicly in the material we read.
  • We have not tested the new settings, and Signal says Android coverage depends on the device maker's build.
  • The Muse findings come from Reuters' testing with Signal and AI Now; Reuters said it could not establish whether some of Muse's odd outputs reported by Meta staff were real leaks or made-up data.

Sources#

Prior Coverage

Earlier Times of AI reporting on this thread.

Scroll to continue reading