On 20 November 2026 Signal said it now blocks screenshots by default on iOS 27, and on Android when it detects an AI assistant reading the screen, to keep Siri AI, Gemini and similar agents out of chats. Signal admits the fix is partial and cannot be applied on macOS. For IT and security teams the lesson is wider than one messenger: an assistant that reads the screen, notifications or file system sits outside every app-level control you rely on, so it needs its own inventory, policy and exclusion list.
This article is general information for IT, security and compliance teams. It is not legal advice; product behaviour is as described on vendor and publisher pages read on 22 November 2026.
What happened
Signal president Meredith Whittaker published a post on the company blog on Friday 20 November titled “Signal is Serious About Protecting Your Privacy”. It announces two changes. On iOS 27 the app prevents screenshots. On Android, where Signal says it has no reliable way to cover every vendor’s build, it instead watches for an AI assistant trying to read the screen and switches on its existing “Screen security” setting when that happens.
Signal points to Google’s Gemini features and to the onscreen awareness capability of Apple’s Siri AI, and compares both to Microsoft Recall, which it blocked on Windows 11 in May 2025 by setting a DRM flag on its own window. Whittaker writes: “This is an existential threat to Signal, and our ability to responsibly uphold our mission.”
Three caveats matter more to practitioners than the headline:
- Users can turn it off. Signal says doing so may expose their content, and their contacts’ content, to OS agents and would “break our privacy guarantees”. The Hacker News gives the path as Settings > Privacy > App Security, with a “Block Siri AI” toggle on iOS and “Screen security” on Android.
- It only addresses screen capture. Signal says blocking screenshots does nothing about agents that reach content already on the device through the file system.
- macOS is not covered. Signal says it cannot reliably block screenshots there, and for now only suggests that macOS 27 users add Signal to the Excluded Apps list under System Settings > Siri.
Signal also says the platforms do not document how their agents reach application data. It asks Apple, Google and Microsoft for two things: a simple, documented way for a developer to opt an app out of AI data access, and proper technical specifications for how the agents handle data.
What the other sources add
The trade press confirms the mechanics. TechCrunch, CyberScoop, ZDNET and The Hacker News report the same core facts. The Register adds that agents typically run optical character recognition over captured screens, and places the move in a wider pattern: it lists Amazon, United Airlines and several large US banks and retailers among companies that have put up barriers to third-party agents.
The Meta Muse reporting shows the problem is not limited to screenshots. Signal’s post refers to third-party agents “like Meta’s Muse” using accessibility permissions as “a skeleton key”. We could not open the underlying Reuters investigation of 18 November (the page returned an access error), so what follows is as relayed by ZDNET and The Hacker News. They report that testing by Reuters and the AI Now Institute found Muse read Signal message text on Android through its access to push notifications, and on a Mac by using accessibility and screenshot tools, in both cases relying on broad permissions granted at setup rather than any Signal-specific connection. The Decoder reports Meta’s position that this happened only with explicit user permission.
Apple is already tightening macOS. An Apple developer notice dated 2 October 2026 says some apps use Full Disk Access in ways that expose files, mail and messages, and adds: “As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially.” A second notice on 16 November says macOS 27.3 beta 2 splits the old Device Control and Data Access permission into two settings (System & App Protection Override, and Accessibility), encourages per-app and per-folder grants, and shows an indicator when one app is controlling another. Existing grants stay in effect.
User-side exclusions exist, unevenly. The Electronic Frontier Foundation reported in October that iOS 27.2 added an “Exclude Apps” setting for Siri AI, which also disables onscreen awareness for the excluded app. Its verdict: “The features not working shows that the new setting is effective.” The EFF notes that Gemini has no per-app switch for screen context. Google’s help page shows the control is global: under Screen context, users turn “Use text from screen” and “Use screenshot” on or off.
Windows is the most manageable of the three. Microsoft’s Recall management page states that Recall is disabled and removed by default on managed commercial devices, that admins cannot switch on snapshot saving for users, and that app filtering and data loss prevention settings are available.
What search trends show
We could not obtain Google Trends interest-over-time data for terms tied to this story; the Explore request was blocked in our environment. As a fallback we checked Google’s daily trending lists for ten countries (US, UK, Germany, India, Türkiye, Japan, Brazil, Australia, Canada, France) on 22 November. None of the listed searches related to Signal, Siri AI, Gemini or Meta Muse. So far this is a specialist story carried by security and technology outlets, not a mass-interest one.
Why this matters for IT and security teams
Most enterprise data protection assumes the application is the boundary. You encrypt in transit, control sign-in and apply app protection policies. An OS-level or broadly permissioned agent ignores that boundary. It reads what the authorised user sees, after decryption, and moves it somewhere the application owner does not control.
Four practical consequences follow.
- Screen, notification and accessibility access are now data-loss channels. In the Muse reporting, no screenshot was needed on Android because notification access was enough. Blocking window capture leaves notifications, the file system and accessibility APIs open.
- Consent given by one person exposes others. When an employee enables an assistant, messages from clients and colleagues in that conversation are swept in too. Signal makes this point about contacts; it applies equally to a deal channel or a customer chat.
- App-level opt-outs are not a platform feature yet. Signal uses a screenshot block and, on Windows, a DRM flag, because nothing better is offered. Android’s documentation describes FLAG_SECURE as a way to stop screenshots of a window; it is a capture control, not an AI opt-out. Teams building internal apps are in the same position as Signal.
- Controls differ sharply by platform. Windows gives admins Recall policy, Apple offers a user-facing exclusion list, and Gemini’s screen context setting is global. One written rule needs three technical implementations, and on unmanaged personal phones it may have none.
Platform by platform: what is and is not covered
| Platform | What Signal does | Stated gap | Lever the sources describe |
|---|---|---|---|
| iOS 27 | Blocks screenshots in the app by default | User can disable; file system access not addressed | “Exclude Apps” for Siri AI from iOS 27.2 (EFF) |
| Android | Enables “Screen security” when an AI assistant tries to read the screen | Detection is not reliable on every vendor build; notifications are a separate route | Gemini Screen context toggles, global only (Google help, EFF) |
| Windows 11 | Screen security on by default since May 2025 | May interfere with screen readers and legitimate screenshots | Recall removed by default on managed devices; filtering and DLP policies (Microsoft) |
| macOS 27 | No screenshot block available | Signal says the APIs do not allow it | Excluded Apps under Siri settings; tighter permissions in macOS 27.3 beta (Apple) |
A practical checklist
- Inventory agents, not just apps. List every assistant on corporate and BYOD devices that can read the screen, notifications, files or accessibility data, including those built into the OS.
- Record permissions. Accessibility, notification access, Full Disk Access and screen capture deserve the scrutiny you give admin rights.
- Decide where content goes. For each assistant, document whether processing is on device or on a vendor server, and what the vendor says about retention.
- Set exclusions for sensitive apps. Use the Siri AI exclusion list, the Gemini screen context setting and Recall policies where your management tooling supports them, and test that they hold.
- Do not instruct staff to disable Signal’s protection for convenience. If screenshots are needed for records, provide an approved export route instead.
- Review notification content. Hiding message previews reduces what a notification-reading agent can collect.
- Harden your own apps. Apply capture protection to screens showing credentials, financial or health data.
- Update the acceptable use policy. State which assistants are approved, on which devices, and that broad device permissions need sign-off.
- Plan accessibility exceptions. Signal acknowledged in 2025 that screen blocking can interfere with assistive software.
What to watch next
- macOS 27.3. The permission changes arrived in beta 2 on 16 November 2026. Apple has not given a release date in the notice we read and says entitlement details will follow “soon”.
- Platform responses to Signal. Signal says it is “cautiously optimistic” about talks with Google and Apple and hopes to share more soon.
- Siri AI in the EU. Apple’s September announcement says Siri AI is not initially available in the EU on iOS, iPadOS and watchOS. Multinational fleets therefore have different exposure by region.
- Follow-up on Meta Muse. The 18 November Reuters investigation is recent, and Meta disputes parts of it. We saw no regulatory action in the sources we read.
- Other privacy-sensitive apps. Signal calls itself the canary. Banking, health and enterprise messaging apps face the same exposure and may ship similar blocks.
Signal has not solved the problem, and says so. Until the operating systems offer a real opt-out, control has to come from device policy, permission hygiene and a clear list of which agents are allowed near which data.
Sources
- Signal: Signal is Serious About Protecting Your Privacy (Meredith Whittaker, 20 Nov 2026)
- Signal: By Default, Signal Doesn’t Recall (21 May 2025)
- TechCrunch: Signal restricts mobile screenshots as it confronts ‘existential threat’ of AI surveillance
- The Register: Signal slams door on prying AI agents to protect privacy
- The Hacker News: Signal Blocks AI Agents From Reading Private Chats via Screenshots on Android and iOS
- CyberScoop: Signal blocks Apple, Google AI screenshotting in response to ‘existential threat’ to privacy app
- ZDNET: Signal to restrict screenshots over AI agent concerns, again
- ZDNET: report on the Reuters investigation into Meta’s Muse agent
- The Decoder: Signal blocks screenshots on iOS 27 and Android to stop AI agents from reading private chats (summary paragraphs only)
- EFF: Victory! You Can Now Exclude Apps from Siri AI
- Apple Developer: Updates to Full Disk Access in macOS (2 Oct 2026)
- Apple Developer: Updates to Device Control and Data Access, and Full Disk Access in macOS (16 Nov 2026)
- Apple Newsroom: Siri AI announcement (September 2026)
- Google Gemini Apps Help: Use screen actions to chat with the Gemini mobile app
- Microsoft Learn: Manage Recall for Windows clients
- Android Developers: Secure sensitive activities (FLAG_SECURE)
- Reuters investigation into Meta Muse (18 Nov 2026): we could not open it (access error); referenced only via ZDNET and The Hacker News.