# Signal Restricts Screenshots to Protect Chats from On-Device AI Assistants (Siri AI, Gemini, Windows Recall)

> Signal is restricting screenshots on iOS 27 and some Android devices so built-in AI assistants such as Apple's Siri AI and Google Gemini cannot read private chats from the screen. End-to-end encryption is not broken; an assistant with screen access can read the decrypted text the app must display, and Signal cannot control how a captured copy is handled.

- **Published:** 2026-11-20T10:42:43Z
- **Last reviewed:** 2026-11-20T10:42:43Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-6568
- **Card image:** https://intel.threadlinqs.com/og/threat/TL-2026-6568.png
- **ID:** TL-2026-6568
- **Severity:** LOW
- **Category:** THREAT_INTEL
- **Status:** MONITORING
- **Detections:** 9 (1 shown free below) · **IOCs:** 15 (1 shown free below)

## Description

This is a privacy and exposure-surface item, not a vulnerability or an active campaign. No CVE, CVSS score, threat actor, malware, network IOC or confirmed exploitation is stated in any source.

According to Cyber Security News (Guru Baran, 2026-11-20), Signal is restricting screenshots on iOS 27 and some Android devices to stop built-in AI assistants from reading private chats through screen capture. End-to-end encryption protects messages in transit, but Signal must show readable text once a chat is opened, so an assistant that can read the screen sees the same text the recipient sees without breaking any encryption. On iOS 27, Signal's 'Block Siri AI' setting prevents Siri from scanning Signal content and blocks screenshots inside the app and in the App Switcher. On Android, Signal automatically enables 'Screen security' when it detects an assistant seeking screen access; Signal says detection varies across Android versions, so protection is not uniform. Users can turn the protection off in Signal's privacy settings, but doing so can expose messages from other people in the chat. Apple says Siri's onscreen awareness lets it answer questions about or act on what the user is viewing. Google's Gemini help page says some screen actions upload visible content to give the assistant context. The article states neither assistant routinely records every Signal chat; the residual risk is that once an assistant captures a message, Signal cannot control how that copy is handled.

The EFF documented the iOS side in two posts. On 2026-09-18 (Thorin Klosowski) it described Siri AI at iOS 27 launch: it searches Apple apps by default, third-party app content depends on the developer indexing it via App Intents, processing runs on-device or in Apple's Private Cloud Compute with no immediate visual indication of which, and at that time there was no way for users or developers to block onscreen awareness. On 2026-10-28 EFF reported that iOS 27.2 added Settings > Siri > Excluded Apps, so that an excluded app's contents are inaccessible to Siri AI, including through onscreen awareness; for Signal this means Siri cannot be invoked to discern information in the app. On Android, EFF says control is less straightforward: Signal does not support Gemini's app-access feature, so it does not appear under Gemini's Connected Apps, but Gemini can still read notifications through the 'Device Assistance' connected app and can use screen context unless 'Use screen and app data' is turned off under Device controls > Screen context.

Signal's own earlier precedent: on 2025-05-21 Signal (Joshua Lund) announced that 'Screen security' is on by default in Signal Desktop for Windows 11. The setting sets the Windows DRM flag on the window so screenshots show nothing, which keeps chats out of Microsoft Recall's searchable history; Signal called Microsoft's lack of granular developer controls 'a glaring omission'. Recall was first announced 2024-05-20, returned to Windows Insiders in April 2025 (Build 26100.3902) and was made opt-in with added local protections. Brave (v1.81, announced 2025-07-23, scheduled for 2025-08-05) took a narrower approach by marking tabs private to Windows so only Recall is blocked while ordinary screenshots keep working; Brave argued Signal's DRM flag also blocks accessibility tools.

Related: Cyber Security News (2026-11-19, citing a Reuters report of 2026-11-18) reports Meta's Muse agent read Signal message text on Android through notifications and on macOS through accessibility and screenshot tools without breaking Signal's encryption; it was blocked on an iPhone during testing.

Limits stated in the source: screenshot blocking does not control what a notification displays, permissions granted to other apps, or content already saved on the device; Signal says it cannot offer the same protection on macOS. Signal is asking Apple, Google and Microsoft for a clear developer opt-out from AI access to screens, files and notifications and for better documentation of what assistants can read and where that data goes.

Defender relevance: treat OS-level AI assistants as a data-exposure surface for decrypted content; review MDM/endpoint policy for assistant screen access, notification previews, accessibility and screen-recording permissions, and screen-capture features on devices handling sensitive communications. Detection content is necessarily speculative because there is no malicious artifact.

## MITRE ATT&CK

- T1113 Screen Capture
- T1513 Screen Capture
- T1005 Data from Local System
- T1533 Data from Local System
- T1517 Access Notifications
- T1517 Access Notifications
- T1567 Exfiltration Over Web Service

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**Microsoft Recall Snapshot Store (ukg.db / CoreAIPlatform) Staged or Archived Outside the Recall Service** (`TL-2026-6568-SPL-002`)

Detects the Microsoft Recall searchable snapshot database ukg.db being created, copied or renamed outside the CoreAIPlatform directory (Sysmon EventCode 11), or command lines that copy, archive or query the CoreAIPlatform store with robocopy, xcopy, Compress-Archive, 7z, rar or sqlite3 (EventCode 1). This indicates local collection and staging of captured screen content (T1005, T1074.001) and archiving (T1560.001) that may include decrypted Signal chat text (TL-2026-6568). Legitimate Recall writes occur inside CoreAIPlatform by AIXHost.exe and are excluded.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1005, T1074.001, T1560.001
- **Data sources:** sysmon, microsoft_defender_for_endpoint
- **Severity:** medium
- **Confidence:** medium

### Splunk SPL

```spl
index=sysmon sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational" ((EventCode=11 TargetFilename="*ukg.db*" NOT TargetFilename="*CoreAIPlatform*") OR (EventCode=1 CommandLine="*CoreAIPlatform*" (CommandLine="*robocopy*" OR CommandLine="*xcopy*" OR CommandLine="*copy*" OR CommandLine="*Compress-Archive*" OR CommandLine="*7z*" OR CommandLine="*rar*" OR CommandLine="*sqlite3*"))) NOT Image="*AIXHost.exe" | eval signal_type=if(EventCode=11,"recall_db_staged_outside_store","recall_store_touched_by_cmdline"), detail=coalesce(TargetFilename, CommandLine) | stats count min(_time) as first_seen values(Image) as process values(User) as user values(detail) as detail by host, signal_type | convert ctime(first_seen)
```

### Microsoft KQL

```kql
let StagedDb = DeviceFileEvents
| where Timestamp > ago(24h)
| where ActionType in ("FileCreated", "FileRenamed", "FileModified")
| where FileName startswith "ukg.db"
| where FolderPath !has "CoreAIPlatform"
| project Timestamp, DeviceName, Actor = InitiatingProcessAccountName, Process = InitiatingProcessFileName, Detail = strcat(FolderPath, " :: ", FileName), SignalType = "recall_db_staged_outside_store";
let StoreAccess = DeviceProcessEvents
| where Timestamp > ago(24h)
| where ProcessCommandLine has "CoreAIPlatform"
| where ProcessCommandLine has_any ("robocopy", "xcopy", "copy", "Compress-Archive", "7z", "rar", "sqlite3")
| where FileName !~ "AIXHost.exe"
| project Timestamp, DeviceName, Actor = AccountName, Process = FileName, Detail = ProcessCommandLine, SignalType = "recall_store_touched_by_cmdline";
union StagedDb, StoreAccess
| order by Timestamp desc
```

### Sigma

```yaml
title: Microsoft Recall Snapshot Database Staged Outside CoreAIPlatform
id: b4d92f68-1a7c-4e35-9c80-2e6f51a7d0b3
status: experimental
description: Detects creation of a ukg.db Recall snapshot database file outside the CoreAIPlatform directory, indicating staging of captured screen content that may contain decrypted Signal chats (TL-2026-6568).
references:
    - https://cybersecuritynews.com/signal-restricts-screenshots/
    - https://signal.org/blog/signal-doesnt-recall/
author: Threadlinqs Team
date: 2026-11-20
tags:
    - attack.collection
    - attack.t1005
    - attack.t1074.001
logsource:
    category: file_event
    product: windows
detection:
    selection:
        TargetFilename|contains: 'ukg.db'
    filter_recall:
        TargetFilename|contains: 'CoreAIPlatform'
    condition: selection and not filter_recall
falsepositives:
    - Backup or forensic tooling copying the Recall store
    - Recall data migration performed by IT
level: medium
```

**Known false positives:**

- Backup or forensic tooling copying the Recall snapshot store
- IT-driven Recall data migration or troubleshooting

**Tuning notes:** Add approved backup/forensics processes to an exclusion list on Image. A renamed copy will evade the filename match; pair with file-hash or size-based hunting if needed. The generic 'copy' substring is intentionally broad on the command-line branch: restrict it to your sensitive-communications host group if noisy.

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- macOS Silent Screen Capture or Accessibility Scripting Against Signal by Non-Interactive Parent (AI Agent Tooling) (locked: unlock with TL Creds)
- PowerShell Scripted Screen Capture via CopyFromScreen / Get-TimedScreenshot / BitBlt (locked: unlock with TL Creds)
- Signal Desktop Data Files Referenced in Upload-Capable Command Line (locked: unlock with TL Creds)
- Android Notification Listener or Accessibility Service Granted to Non-Allowlisted App (Signal Exposure) (locked: unlock with TL Creds)
- Signal [indicator] / [indicator] Protection Disabled or Signal Not in Siri Excluded Apps (locked: unlock with TL Creds)
- Gemini Screen Context (Use screen and app data) Enabled on Managed Android Device (locked: unlock with TL Creds)
- Microsoft Recall Enabled via WindowsAI Policy or Optional Feature (Signal Chat Exposure Surface) (locked: unlock with TL Creds)
- Screen Capture / Recall Component Outbound Connection to Non-Microsoft Destination (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sample indicators (1 of 15)

Defanged, TLP:CLEAR.

- Domain: `eff[.]org`

Full set of 15 indicators unlocks with TL Creds: https://threadlinqs.com/tl-creds

## Sources

- [Signal Restricts Screenshots to protect User Privacy from AI assistants](https://cybersecuritynews.com/signal-restricts-screenshots/)
- [EFF: Victory! You Can Now Exclude Apps From Siri AI](https://www.eff.org/deeplinks/2026/10/victory-you-can-now-exclude-apps-siri-ai)
- [EFF: How to Limit What Apple's New Siri AI Can Access in iOS 27](https://www.eff.org/deeplinks/2026/09/how-limit-what-apples-new-siri-ai-can-access-ios-27)
- [Signal Support: Screen Security](https://support.signal.org/hc/en-us/articles/360043469312-Screen-Security)
- [Signal Blog: By Default, Signal Doesn't Recall](https://signal.org/blog/signal-doesnt-recall/)
- [Meta's Muse reads encrypted messages (Cyber Security News)](https://cybersecuritynews.com/muse-reads-encrypted-messages/)
- [Brave browser blocks Microsoft Recall (Cyber Security News)](https://cybersecuritynews.com/brave-browser-blocks-microsoft-recall/)
- [The Register: Brave browser blocks privacy-busting Microsoft Recall](https://www.theregister.com/2025/07/23/brave_browse_block_microsoft_recall/)
- [Brave: Blocking Microsoft Recall by default](https://brave.com/privacy-updates/35-block-recall/)
- [SecurityWeek: Signal Adds Screenshot Blocker to Thwart Windows Recall](https://www.securityweek.com/signal-adds-screenshot-blocker-to-thwart-windows-recall/)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple and Gold in full, 2 calls a day on a free account): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-6568
