Signal, the encrypted messaging app, is rolling out a setting on iOS and Android that blocks screenshots and prevents operating-system AI assistants from reading what's displayed in the app. The change is a direct response to a new category of AI features that scan screen content to provide context-aware help โ€” and a sign of where the next privacy battle in consumer tech is headed.

The feature works as a toggle: users can disable screenshots within Signal, which also blocks the kind of screen analysis that powers tools like Apple's Siri and Google's Gemini when they're granted system-wide visual access. These AI assistants are increasingly built to look at whatever is on a user's screen โ€” a text message, a photo, a webpage โ€” and use that information to answer questions or take actions, without the user explicitly sharing the content.

Signal has built its reputation on end-to-end encryption, meaning messages are scrambled in transit so that not even Signal itself can read them. But encryption only protects data while it's moving or stored. Once a message is decrypted and displayed on a user's screen, an AI agent with screen-reading permissions can see it just as a human eye would โ€” a gap that encryption was never designed to close.

This isn't the first time an app has restricted screenshots for privacy or security reasons. Banking apps have long disabled screenshots to prevent account details from being captured, and apps like Snapchat built entire features around notifying users when a screenshot is taken. What's new here is that the target isn't other people โ€” it's the AI running on your own phone.

Why it matters

The move lands in the middle of a broader industry push toward agentic AI โ€” assistants that don't just answer questions but actively observe and act across apps. Microsoft's Recall feature, which periodically captures screenshots of a user's activity to build a searchable history, drew enough backlash over privacy concerns that the company delayed and scaled back its rollout. Apple and Google have been more cautious in how they frame screen-reading AI, but the underlying capability โ€” software that can see what's on your display โ€” is becoming a standard feature of modern operating systems, not a niche one.

Signal's decision puts a spotlight on a tension that most app developers haven't had to confront before: when the AI threat isn't a hacker or a data broker, but the phone's own operating system.

What this means for small businesses

Any business using messaging apps, CRM software, or financial tools on a phone or laptop with AI assistants enabled should treat this as a prompt to check what those assistants can actually see. Features marketed as convenience โ€” summarizing emails, answering questions about what's on screen โ€” often require broad permissions that extend to every app on the device, not just the ones you'd expect.

This is especially relevant for businesses handling client data, financial records, or anything covered by confidentiality agreements or regulations like HIPAA. An AI assistant that can read a screen doesn't distinguish between a casual text and a client's banking details displayed in an accounting app.

The practical step this week: go into your phone's AI assistant settings โ€” Siri, Gemini, or Copilot โ€” and review what permissions are granted for screen access or app visibility. Most operating systems allow these features to be limited on a per-app basis, even without a developer-side block like Signal's.

What to watch

Keep an eye on whether other privacy-focused apps โ€” Proton Mail, WhatsApp, password managers โ€” introduce similar screen-blocking controls, and whether Apple or Google respond by changing default permissions for agentic AI features. Also worth tracking: any regulatory guidance on AI screen-reading, which so far has been largely unaddressed by existing privacy law.

The bottom line

Signal's move doesn't fix the underlying issue industry-wide โ€” it only protects what happens inside one app. Businesses relying on AI-equipped devices for sensitive work should treat screen-reading permissions as a setting worth auditing, not a default to leave untouched.