Back to Intelligence

Signal Blocks Siri and Gemini AI From Capturing Chats on iOS and Android — What Defenders Must Do Now

SA
Security Arsenal Team
November 22, 2026
8 min read

Signal has announced a new set of platform-level protections designed to keep conversations out of the reach of on-device AI assistants and screenshot mechanisms. On iOS 27, Signal will block screenshots of chats entirely, and on supported Android devices, the app will prevent AI assistants — specifically naming Apple's Siri and Google's Gemini — from capturing conversation content.

This is not a hypothetical privacy concern. Over the past 18 months, the security community has watched operating system vendors aggressively integrate large language model assistants at the OS level. These assistants increasingly have the ability to read screen content, summarize conversations, index notifications, and — critically — transmit that context to cloud infrastructure for processing. For any organization that selected Signal precisely because it offers end-to-end encryption with a minimal data footprint, the arrival of OS-level AI that can read everything on screen represents a fundamental bypass of the encryption model. The message content is encrypted in transit and at rest in Signal's data store — but none of that matters if an AI assistant captures the decrypted plaintext rendered on the display.

Signal's move acknowledges an uncomfortable truth for defenders: the endpoint rendering layer is now a first-class data exfiltration surface. If your threat model includes nation-state actors, industrial espionage, insider threats, or simply regulatory exposure under HIPAA, PCI-DSS, or attorney-client privilege, you need to understand what changed and what you must enforce.

Technical Analysis

What Signal Is Changing

Signal's announcement covers two distinct platform protections:

iOS 27 — Screenshot Blocking. Apple introduced a new screen-capture prevention API in iOS 27 that allows apps to mark their UI windows as protected from screenshots and screen recording. Signal is adopting this API to block capture of chat content at the OS compositor level. This is materially stronger than the legacy isScreenCaptured detection approach, which could only detect a screenshot after the fact — the new API prevents the capture from containing the sensitive content in the first place.

Android — AI Assistant Capture Prevention. On supported Android devices, Signal will leverage Android's privacy flags (the app has long supported the FLAG_SECURE window flag for its Screen Security feature) and newer platform controls that restrict AI assistant frameworks — including Gemini's screen-context features — from ingesting conversation content. This addresses features like Gemini's ability to read and summarize on-screen content, circle-to-search style capture, and assistant-driven screenshot analysis.

Why This Matters From a Defender's Perspective

The attack surface here is not a vulnerability in Signal — Signal's cryptography remains intact. The risk is plaintext exposure at the presentation layer, via three channels:

  1. OS-integrated AI assistants that capture screen content to answer contextual queries, with that content potentially leaving the device for cloud inference.
  2. Screenshot and screen-recording capabilities available to the user, other apps, MDM tooling, or malicious processes with accessibility privileges.
  3. Accessibility services abuse — a well-documented Android malware technique (used by families like SpyNote, BrasDex, and numerous bankers) that reads screen content of any app that does not set secure window flags.

Signal's protections close channels 1 and 2 on supported platforms, but only if the features are actually enabled and the devices are running supported OS versions.

Scope and Limitations — Read This Carefully

  • iOS 27 is required for the screenshot-blocking capability. Devices on older iOS versions get no new protection. Signal cannot retroactively harden an OS that lacks the API.
  • Android support is device- and version-dependent. Signal's announcement specifies "supported Android devices" — meaning OEMs that ship the relevant AI-privacy controls. Fleet heterogeneity matters here; your Samsung devices may behave differently from your Pixels.
  • Screenshots of notifications are a separate concern. Chat content rendered in lock-screen notifications is outside Signal's chat window and may still be capturable. Notification content visibility must be disabled separately in Signal settings.
  • A camera pointed at the screen defeats all of this. These controls raise the bar against software-based capture; they do nothing against physical capture. Adjust your threat model accordingly.
  • No CVE is associated with this announcement. This is a proactive platform hardening measure, not a response to a disclosed vulnerability or observed exploitation. However, the technique it defends against — AI assistant screen-context harvesting — is actively shipping in consumer OS builds today, making the exposure current and real.

Executive Takeaways

This is a platform/privacy feature announcement rather than an exploit or malware campaign, so the value here is in organizational action, not detection rules. These are the steps I'd put in front of a CISO this week:

  1. Update your approved-messaging policy now. If Signal (or any E2EE messenger) is sanctioned for sensitive communications — incident response coordination, M&A discussions, whistleblower channels, clinical or legal communications — your acceptable use policy must account for the fact that OS-level AI assistants can read screen content independent of app encryption. Document which device/OS combinations are approved.

  2. Mandate minimum OS versions via MDM. Push iOS 27 and current Android builds as the floor for any device used for sensitive communications. Devices that cannot run these versions should be excluded from sensitive comms workflows. Enforce this through your MDM compliance policies rather than trusting users to update.

  3. Disable AI assistant screen-context features on managed devices. On Android enterprise devices, use your EMM to restrict Gemini/assistant screen capture and summarization features. On iOS, review and restrict Apple Intelligence's screen-awareness capabilities where your MDM framework exposes those controls. Do not assume Signal's in-app protection covers every assistant invocation path — defense in depth means disabling the capture capability at the OS policy layer too.

  4. Enable Signal's full privacy stack, not just the new features. Verify that Screen Security (Android), screen lock, disappearing messages, and notification content suppression are enabled on all devices used for sensitive communications. Signal's new AI-blocking controls complement — they do not replace — these existing settings.

  5. Reassess your mobile threat model for accessibility abuse. The same screen-reading capability that powers legitimate AI assistants is what Android malware abuses via accessibility services. Ensure your mobile threat defense (MTD) tooling flags sideloaded apps requesting accessibility permissions, and audit existing accessibility grants on managed devices.

  6. Brief leadership and legal/compliance stakeholders. If your organization relies on Signal for privileged or regulated communications, the emergence of OS-level AI screen capture is a material change to your data-flow model. Privacy officers and counsel should understand where decrypted message content can now flow (assistant cloud inference pipelines) so data handling attestations remain accurate.

Remediation and Hardening Steps

There is no patch to apply — this is a configuration and policy remediation effort. Execute the following:

On the device (per-user or MDM-enforced):

  • Update Signal to the latest version from the App Store / Google Play to receive the new protections as they roll out.
  • Update devices to iOS 27 (or later) and the latest supported Android release.
  • In Signal: enable Screen Security (Android: Settings → Privacy → Screen Security), Screen Lock, and set Notifications → Show to "No name or message" or "Name only."
  • Disable assistant screen-context features: on Android, review Gemini's "Use screen context" settings; on iOS, review Siri/Apple Intelligence screen-awareness permissions.

At the organization level:

  • Add OS-version compliance gates to MDM for any device in the sensitive-communications device group.
  • Use Android Enterprise policies to restrict assistant invocation and screen capture APIs on work profiles.
  • Update your data classification policy: define which classifications of information may be discussed over mobile E2EE messaging at all, and on which device/OS combinations.
  • Monitor Signal's official release notes and the platform vendors' enterprise documentation for the exact API and MDM controls as they ship — the iOS 27 capture-prevention API in particular will likely be exposed as an MDM-manageable restriction, and you'll want to enforce it fleet-wide rather than relying on per-app adoption.

Vendor references:

The Bottom Line

Signal's move is a bellwether. End-to-end encryption solved the in-transit problem years ago; the industry is now confronting the fact that the biggest remaining plaintext exposure is the screen itself — and OS vendors just handed AI assistants a front-row seat to it. Expect every privacy-focused app to follow Signal's lead, and expect adversaries to shift toward exactly the channels these controls leave open: physical capture, compromised OS builds, and accessibility-layer abuse. Treat mobile presentation-layer privacy as a managed control now, with OS minimums, MDM enforcement, and policy language to match. The organizations that get ahead of this will not be the ones scrambling when the first "AI assistant leaked the deal" incident hits the news.

Related Resources

Security Arsenal Managed SOC Services AlertMonitor Platform Book a SOC Assessment soc-mdr Intel Hub

Is your security operations ready?

Get a free SOC assessment or see how AlertMonitor cuts through alert noise with automated triage.