Signal said on Friday, November 20, that it now blocks screenshots by default in its apps on iOS 27 and, on Android, whenever it detects an AI assistant trying to read the screen. The goal is to stop agents built into the operating system, such as Apple’s Siri AI with onscreen awareness and Google’s Gemini Intelligence, from capturing decrypted messages. Our view: Signal is right about the problem, but a screenshot ban is a workaround, and the real fix belongs to Apple, Google and Microsoft. Operating systems need a clear, enforceable way for apps to tell AI agents “not here.”

What Signal changed and why

In a blog post, Signal president Meredith Whittaker wrote that OS-level agents “scan” what is on the screen so they can answer questions and act for the user. Once Signal messages are captured that way, they can be stored and sent to “whatever local database or remote server” the agent uses, outside Signal’s control. Whittaker called this an “existential threat” to Signal’s ability to guarantee privacy.

Signal has done this before. Last year it used the same screenshot-blocking technique to keep Microsoft’s Recall feature from recording chats on Windows 11. Users who want screenshots back can turn the protection off in settings, though Signal warns that doing so may expose their and their contacts’ messages to these agents. The company also admits the fix is incomplete: it cannot be implemented on macOS, and agents can still reach app data through other routes such as notifications or the file system. The Decoder notes that Meta’s Muse agent recently captured Signal content on Android and macOS through broad accessibility and screenshot permissions, which Meta says it did only with explicit user consent.

Why a screenshot ban is the wrong tool

Blocking screenshots is a blunt instrument because it treats every capture as the same. A person saving a receipt for a group dinner and an agent silently feeding a chat history into a cloud model look identical to the app. To stop the second, Signal has to break the first for everyone.

A person holding a smartphone and looking at its screen
Agents built into phone operating systems can read whatever appears on screen, including decrypted messages; the photo is illustrative. Photo: Océanos y dados / Wikimedia Commons, CC0

It also turns a design question into a cat-and-mouse game. Agents increasingly operate computers directly; this week we covered how the Gemini app for Mac can now click and type in other apps from a background window. Every new agent surface, whether accessibility services, screen recording or notification access, is another hole an app developer must find and plug, usually after the feature has shipped. Developers cannot secure what platforms do not document, and Signal says it lacks basic information about how these agents access app data.

What platforms should offer instead

The operating system already knows how to handle sensitive surfaces. Password fields are excluded from many capture and autofill features, and apps can mark windows as protected from screen recording. Extending that model to agents would not be hard in principle:

  • An app-level opt-out. A flag that tells every built-in or third-party agent that an app or a specific view is off limits, enforced by the OS rather than by each agent’s goodwill.
  • Per-app consent for agents. The same kind of prompt users see for camera or location, so a person can let an agent use their calendar but not their messenger.
  • Visible indicators and logs. A clear signal when an agent is reading the screen, and a record of what it took, so users and developers can audit it.
  • Accurate documentation. Published descriptions of what each agent captures, where it is processed and how long it is kept.

None of this stops users from choosing to share a conversation with an assistant. It moves that choice from a hidden system setting to an explicit decision, and it gives app makers a standard tool instead of a hack. The same logic applies to security teams worried about prompt injection through on-screen content: an agent that reads everything also ingests everything, including hostile text.

The counterargument, and why it falls short

Apple and Google can fairly say that users asked for assistants that understand context, and that a screen-reading agent is only as invasive as the person allows. Some users will want Siri to summarize a long group chat. But Signal’s case is not about one user’s consent. A conversation has several participants, and only one of them is choosing to let an agent read it. End-to-end encryption exists to protect everyone in that chat, and an OS feature that quietly bypasses it for one participant changes the deal for all of them.

Our explainer on sandboxing AI agents argues that agents should get the minimum access a task needs. Operating systems are now doing the opposite by default, granting agents broad visibility first and leaving apps to claw it back.

What to watch next

  • Whether Apple, Google or Microsoft announce developer controls for excluding apps from agent access.
  • Other privacy-focused apps, such as password managers and banking apps, adopting similar blocks.
  • Regulators in the EU and elsewhere examining OS-level agents under data protection law.
  • Signal finding a way to extend protection to macOS.

Key takeaways

  • Signal now blocks screenshots by default on iOS 27 and on Android when it detects AI assistants reading the screen.
  • The company says OS-level agents undermine its end-to-end encryption guarantees and calls the threat existential.
  • Screenshot blocking is incomplete and penalizes ordinary users, and it cannot work on macOS.
  • We think operating systems should offer an enforceable app opt-out, per-app agent consent and clear logs.

Frequently asked questions

Why is Signal blocking screenshots?

Signal says AI agents built into iOS and Android capture on-screen content, including decrypted messages, and can store or send it elsewhere. Blocking screenshots is its way of preventing that capture.

Can I turn Signal’s screenshot blocking off?

Yes. The protection can be disabled in Signal’s settings, but Signal warns this may expose your messages and your contacts’ messages to AI agents on the device.

Does Signal’s protection work on Mac?

No. Signal says operating systems do not give it the tools to reliably block AI capture on macOS without breaking the user experience, so the protection currently applies to iOS 27 and supported Android versions.