What Signal announced
On November 20, 2026, Signal president Meredith Whittaker published a blog post that is blunt about the problem. Signal will "restrict agentic systems built into iOS 27 and on Android versions where we can detect an AI assistant attempting to read your Signal screen."
The mechanism is simple. OS-level AI agents work by reading what is on your display. In Signal's telling, that captures your messages and hands them to systems that may store them in a local database or on a remote server. Signal says this leaves the app with no control over the data. It also affects everyone in a conversation, not just the person whose phone is scanned.
This is not a break in encryption. Signal's own support page says the new protection affects only screen capture and on-device scanning. It does not change how messages are encrypted, who can see them, or any other setting.
Signal names Google's Gemini Intelligence and Apple's Siri AI, with its Onscreen Awareness feature, as the examples driving the change. Reuters, via Investing.com, reports that Signal singled out AI built directly into operating systems because of how deeply it is integrated into devices.
The Windows Recall precedent
Windows readers will recognize the pattern. Signal says it used screenshot blocking last year to stop Microsoft Recall from capturing Signal messages on Windows 11 by default. In a footnote, Signal describes Recall as taking screenshots every few seconds, extracting text with optical character recognition, and building a searchable on-device database. It says this could retain Signal content, including disappearing messages, indefinitely. That is Signal's characterization of Recall, not Microsoft's.
Several outlets mark the mobile move as a repeat of that approach. Cyberscoop describes it as following similar steps taken with Windows 11 last year, now extended to Siri AI's Onscreen Awareness and Gemini Intelligence.
The mobile case is arguably harder than Recall. Recall was a single, identifiable feature. The new problem is a family of assistants, built by different vendors, that behave differently on different devices.
Versions, settings and defaults
Signal's support documentation sets out the requirements:
- Android: Signal 8.35.1 or later.
- iPhone: Signal 8.37 or later, running iOS 27 or later.
The two platforms do not behave the same way.
iPhone
- The "Block Siri AI" setting is on by default.
- It is found at Settings → Privacy → Block Siri AI.
- The options are On, Off, and Off for 10 Minutes. The temporary option turns protection back on automatically after ten minutes, which suits a quick screenshot.
- With protection on, Signal blocks screenshots and screen recordings inside the app. It also keeps Signal content out of the App Switcher and blocks on-device assistants from reading the screen.
- If you attempt a screenshot, Signal shows a message saying it was prevented. An Options button lets you turn protection off, turn it off for ten minutes, or leave it on.
- Signal asks you to confirm before turning protection off, because assistants may then be able to scan your content.
Android
- Signal's existing "Screen security" setting is switched on automatically the first time an AI assistant tries to read Signal's screen. Signal shows a message explaining what happened.
- You can enable it earlier at Settings → Privacy → App security → Screen security.
- It is a plain on/off control.
- With it on, the phone will not take a screenshot inside Signal. It also hides content from the recent-apps list. Enabling it also enables the Hide Screen in App Switcher setting.
- If you disable it by tapping "Turn off screen security" in the prompt, Signal turns it back on after 90 days. If you switch it off manually in Settings, there is no reminder.
Why detect an assistant instead of simply blocking screenshots on Android? Signal's footnote explains that Android varies by vendor. On many devices it cannot reliably detect a screenshot attempt when screenshots are already disabled. It chose to detect AI access attempts rather than leave users with screenshots that silently stop working. The trade-off is that it cannot cover every Android device.
What the protection does not do
Signal itself calls the fix incomplete, and the caveats matter.
- Other access routes. Screenshot blocking does not address agents that read content already on the device through the filesystem. Signal also lists notifications, APIs and other OS frameworks as paths developers need better controls over.
- The other person's phone. Engadget notes that if your contact has switched protections off, blocking screenshots on your end won't stop an AI assistant or agent from reading your messages on the other person's device.
- Third-party agents. Signal points to agents such as Meta's Muse that use accessibility permissions granted at setup. Those permissions were meant to help people use their apps, not to serve as what Signal calls a skeleton key for agents.
- macOS. There is no equivalent protection on Mac. Signal says it cannot reliably block screenshots there and lacks clearly documented APIs. It says the available macOS protections against continuous capture don't cover screenshots, and some APIs are geared toward streaming and DRM-protected media. For macOS 27 users, its only recommendation is to add Signal under System Settings → Siri → Excluded Apps. That guards against proactive Siri AI access and nothing more.
The Muse example
Reuters ties the announcement to its own reporting earlier in the week. It said Meta's Muse agent bypassed its official "Connector" tool to read Signal messages on both an Apple computer and an Android phone. According to Reuters, Muse used broad OS-level permissions granted at setup. On macOS it combined accessibility and screenshot tools to navigate to Signal, capture chats, extract the text and reproduce the messages inside Muse.
Reuters adds that an Apple OS update that week closed that specific pathway. The update also introduced new permissions and warnings aimed at agents with similar capabilities. Meta told Reuters that agents need to see and interact with a user's computer to work on their behalf, and that Muse accesses other apps only through permissions users explicitly grant.
This is a fair counterpoint. Agent vendors can reasonably argue that a user who grants screen and accessibility access has consented. Signal's reply is that users are encouraged to grant such deep permissions during setup and may not anticipate how they will be used.
The Register adds that Apple's macOS 27.3, currently in public beta, tightens controls around Device Control and Data Access and Full Disk Access. The Register says this appears aimed mainly at third-party access rather than Siri itself.
What Signal wants from Microsoft, Apple and Google
Signal's demands are specific:
- Developers should have an easy, documented way to opt an app out of AI data access and processing. That covers screen reading, filesystem access, notifications, APIs and other OS frameworks, without gutting the app's core functionality.
- Operating system makers should publish better documentation and technical specifications on how their AI agents work, especially their data access and processing.
Signal also says confidential computing is not a complete answer, because it still leaves providers in charge of the privacy guarantees. It says it has spent considerable developer time on forensic investigation of Android and iOS because documentation is lacking. It is "cautiously optimistic" about talks with some OS providers, including Google and Apple, but gives no timetable.
Microsoft is named alongside Apple and Google as a company that doesn't offer developers easy restrictions without drastic compromises to the user experience. Reuters reports that Apple, Google and Microsoft did not immediately respond to requests for comment.
Why this matters for Windows and IT readers
- The Recall debate isn't over. The same design tension now runs across all three major platforms. Any Windows app that handles sensitive content faces it, as do Copilot-style features that can see the screen.
- Admins should separate threats. Encryption, screen capture, notifications, filesystem access and accessibility permissions are different attack surfaces. Blocking one doesn't close the others.
- Permissions are the real control point. Reuters' Muse example shows that broad accessibility and screen-capture grants can undo app-level protections. Reviewing which agents hold such permissions on managed devices is a sensible first step. This is general guidance, not something Signal or Reuters prescribes.
- Developers lack a clean opt-out. Until OS vendors supply one, app makers are left with workarounds such as screenshot blocking and heuristic detection.
Practical checklist
- Update Signal to 8.35.1 or later on Android, or 8.37 or later on iPhone running iOS 27.
- On iPhone, confirm Block Siri AI shows On. On Android, check that Screen security is enabled.
- Turn protection off only briefly, using the iPhone 10-minute option where possible, and remember the exposure includes your contacts' messages.
- On a Mac running macOS 27, add Signal to Siri's Excluded Apps list.
- Audit which AI apps have accessibility or screen-recording permissions on your devices.
Bottom line
Signal's move is a useful, narrowly scoped mitigation, not a cure. It shows that app developers now treat OS-level AI assistants as a privacy threat comparable to Recall. Whether Microsoft, Apple and Google will give developers a real opt-out is the open question. For now, Signal's own admission is that the protection is partial.
References
- Signal slams door on prying AI agents to protect privacy The Register · 2026-11-20T20:42:04+00:00
- Signal restricts screenshots to block AI agents from reading messages By Reuters investing.com
- Screen security: blocking Google Gemini and Siri AI – Signal Support support.signal.org