Back

Signal Calls OS-Level AI an 'Existential Threat' and Blocks Screenshots to Fight It

NeutralShare toX (Twitter)Telegram
Signal Calls OS-Level AI an 'Existential Threat' and Blocks Screenshots to Fight It

Key Takeaways

  • Signal is blocking screenshots to stop smartphone AI assistants like Google’s Gemini and Apple’s Siri from scanning and storing private chats.
  • The company’s president Meredith Whittaker called the OS-level AI integrations an “existential threat,” saying captured data lands outside Signal’s encryption.
  • It flagged third-party agents like Meta’s Muse, which previously accessed a journalist’s iMessages without consent.

Signal is moving to wall off its messages from the AI assistants now baked into smartphones, restricting screenshots so that systems like Google’s Gemini and Apple’s Siri can’t quietly scan and store users’ private conversations.

The encrypted-messaging nonprofit said Thursday it will block the latest agentic AI features built into iOS 27 and, on Android, automatically enable a “screen security” setting whenever it detects an AI assistant trying to read a user’s Signal screen.

The move extends a tactic Signal first used last year against Microsoft’s Recall, the Windows feature that captured periodic screenshots; the company blocked it to keep Recall from storing images of Signal messages by default.

The concern, Signal says, is that operating-system AI agents work by “scanning” whatever is on a user’s screen, which means Signal data gets captured and handed to those systems to process.

Once that happens, the content lands in whatever local database or remote server the AI is using, outside Signal’s control and beyond the encryption guarantees that have defined the app for a decade.

Signal president Meredith Whittaker framed it starkly, calling the OS-level AI integrations “an existential threat” to the app’s mission and arguing that Apple, Google and Microsoft are reshaping the environment every app runs in “in a way that’s making it increasingly hostile to privacy.”

Users who still want to take screenshots can disable the protection in Signal’s settings, though the company warns that doing so could expose their messages, and their contacts’ messages, to the AI agents.

Signal also conceded the fix is incomplete: It doesn’t stop sneakier methods of access, such as pulling data straight from a device’s filesystem, and it can’t yet be implemented on macOS, where Signal says Apple lacks the documented tools to block screenshots reliably.

The privacy-focused company also called on the three OS giants to give developers a simple, well-documented way to opt apps out of AI data access, and to publish clear specifications on how their agents actually handle data, saying it has spent months on “forensic investigations” just to understand what these systems do.

Signal also flagged third-party agents like Meta’s Muse, warning they exploit accessibility permissions meant to help people, not to serve as “a skeleton key” for AI.

The worry isn’t hypothetical: A journalist found that Muse had accessed his private iMessages despite his settings and then gave a false explanation of how it knew their contents, an episode that crystallized fears about how much agents can reach once granted deep device permissions.

Signal’s move lands amid a broader reckoning over AI agents accessing data they shouldn’t, from rogue agents breaching systems to banks scrambling to keep autonomous software out of sensitive accounts.

Read the original at Decrypt

More News