Signal Starts Blocking Phone AI Assistants From Reading Its Chats
Signal is now defaulting to screenshot and screen-access blocks on iOS and Android to keep Apple's Siri AI and Google's Gemini from reading private chats, days after a Reuters investigation found Meta's AI agent could pull Signal message text on Android.

Signal said on Friday that it is switching on, by default, a set of settings designed to stop the artificial intelligence assistants now built into phone operating systems from reading the contents of its chats, a response to what the encrypted-messaging app's leadership calls a growing threat from AI agents that can see whatever is on a user's screen.
In a blog post published by Signal, company president Meredith Whittaker said the app will block Apple's Siri AI from reading Signal conversations on iOS 27 through a new "Block Siri AI" setting, and will automatically switch on its existing "Screen security" feature on Android whenever it detects an AI assistant, such as Google's Gemini, attempting to read the screen. The move extends a similar default protection Signal built into its Windows desktop app last year to block Microsoft's Recall feature. Reuters, which first reported the change, said Signal is also prompting users with pop-ups when it detects a screenshot attempt or an AI assistant accessing the screen.
What is changing, and what is not
The new defaults do not amount to a complete fix, and Signal says so explicitly. On Apple's Mac computers, the company says it could not find a reliable way to block screenshots or Siri's "onscreen awareness" feature at all, and its only recommendation for macOS 27 users is to manually add Signal to the Excluded Apps list under System Settings. Even on the platforms where the new protections do work, Signal warns that AI agents can still reach messages and files that are already stored, unencrypted, on a device, through other channels such as accessibility permissions or notification access.
Signal's own history with this kind of blocking goes back to 2025, when it began defaulting to blanking screenshots of its Windows app to defeat Microsoft's Recall feature, which periodically captures and indexes what appears on a user's screen. The underlying worry, Whittaker wrote, is that operating-system vendors are now building AI directly into the software layer that sits beneath every app, giving it a vantage point no previous feature had.
"This is an existential threat to Signal, and our ability to responsibly uphold our mission."
That is how Whittaker described the stakes in the blog post, according to Reuters' account of the announcement. Signal is a nonprofit that does not disclose official user numbers, though Whittaker has previously put its monthly user base at roughly 70 million to 100 million people worldwide.
A feature Apple and Google built, and a breach Meta is explaining
The screen-reading capability Signal is trying to block is a deliberate product feature, not a bug. Apple introduced Siri AI in June, describing "onscreen awareness" that lets the assistant answer questions about whatever is currently displayed. Google's Gemini assistant on Android similarly relies on special permissions, such as notification access, that Android's own developer documentation describes as requiring a user to grant access through a dedicated settings screen rather than a routine prompt, as laid out in Google's permissions guidance for app developers. Apple added a separate "Excluded Apps" control to Siri's settings in October, letting users block the assistant from specific apps altogether, a change the digital-rights group Electronic Frontier Foundation welcomed as a privacy win while cautioning that onscreen awareness remains difficult for any outside developer to fully control.
Signal's announcement follows a Reuters investigation, conducted with Signal's cooperation and the AI Now Institute, that found Meta's "Muse" personal AI agent could retrieve text from Signal conversations on Android by using notification-access permissions, without going through the dedicated "Connector" feature Meta built for that purpose. An equivalent attempt on an iPhone was blocked. A Meta spokesperson told Reuters that Muse "can access other apps only via permissions users explicitly grant" and described the episodes identified by reporters as isolated incidents that the company has since addressed.
Signal's broader argument, repeated across Whittaker's public statements over the past year, is that any AI agent capable of reading a phone's screen effectively bypasses the encryption Signal uses to protect messages in transit, because the content has to be rendered as pixels at some point for a human to read it — and those same pixels are now visible to software running underneath the app.
What Signal is asking for, and what happens next
The company is calling on Apple, Google and Microsoft to go further than the current, piecemeal opt-outs. In the blog post, Signal asked the platform makers for simple, well-documented ways for developers to exclude their apps from AI data collection across every channel an agent might use — screen reading, notifications and file-system access among them — rather than requiring separate settings for each. It also wants the companies to publish clearer technical documentation describing exactly how their AI agents access and process data, arguing that confidential-computing promises from providers do not resolve the underlying question of who controls the privacy guarantee.
Reuters reported that Apple, Google and Microsoft had not responded to requests for comment by the time the story published. Whittaker has described Signal's position as that of "the canary in the coal mine" for a wider privacy problem facing every app developer as AI agents become a standard part of mobile and desktop operating systems, rather than an issue unique to encrypted messaging.
For now, the practical effect for Signal's users is a handful of new toggles and pop-up warnings rather than a structural change to how the app works. Users who want AI assistants to retain screen access to their chats can turn the new protections off again in settings, something Signal's own messaging acknowledges undercuts the goal: a default that can be switched off is not the same as a guarantee. The company says it will keep pushing the three platform makers for stronger, enforceable commitments, and has signaled it expects the fight over where AI agents may and may not look to continue well beyond this week's update.

Anthropic mathematician says he and an AI model may have solved a 167-year-old prime-number puzzle
OpenRouter Pulls 125 Endpoints After Research Finds AI 'Privacy' Promises Were Broken

Thinking Machines releases a laptop-ready AI model built on reused layers
