Signal has started blocking screenshots inside its mobile app, and the target is not nosy contacts. It is the AI assistants that Apple and Google are building into their operating systems. In a blog post published on November 20, 2026, Signal president Meredith Whittaker said the messaging service will restrict “agentic systems built into iOS 27” and, on Android, react whenever it detects an AI assistant trying to read the Signal screen. Her explanation was blunt: OS-level AI integrations are “changing this environment in a way that’s making it increasingly hostile to privacy,” which she called “an existential threat to Signal, and our ability to responsibly uphold our mission.” For anyone who manages company phones, writes mobile apps or advises on AI deployments, the episode is a preview of a fight that is coming to every privacy-sensitive app.
TL;DR
- Signal now blocks screenshots and screen recordings by default on iPhones running iOS 27 to stop Siri AI from scanning chats, and on Android it turns on its Screen security setting the first time it detects an AI assistant reading the screen.
- Users can switch the protection off, but Signal warns that doing so may expose their own and their contacts’ messages to OS-level AI agents.
- Signal says the fix is incomplete, cannot be applied on macOS, and asks Apple, Google and Microsoft for a documented way for apps to opt out of AI data access.

What Signal actually changed
Early summaries of the news described “message sharing restrictions.” That is not quite right. Signal has not limited forwarding or sharing of messages. What it has done is restrict screen capture, because screen capture is how the new assistants see what is inside an app.
According to Signal’s support page, the protection works differently on each platform:
- iPhone (Signal 8.37 or later on iOS 27 or later): a new “Block Siri AI” setting is on by default. It blocks screenshots and screen recordings in the app, hides Signal content in the App Switcher and prevents on-device assistants from reading the screen. Users can leave it on, turn it off, or turn it off for 10 minutes to grab a quick screenshot.
- Android (Signal 8.35.1 or later): the existing Screen security setting switches on automatically the first time an AI assistant tries to read Signal’s screen, and the app shows a message explaining what happened. Users can enable it in advance. If someone turns it off from that prompt, Signal turns it back on after 90 days.
- Encryption is untouched: Signal says the change affects only screen capture and on-device scanning, not how messages are encrypted or who can read them.
The Android design is a compromise. Signal explained that on many Android builds from manufacturers other than Google, it cannot reliably tell when a screenshot was attempted. Rather than have screenshots silently fail, it chose to watch for an assistant reaching for the screen and respond then.
Why screen-reading assistants are a problem for an encrypted app
Signal’s security model rests on end-to-end encryption and an open-source code base that outside researchers can inspect. Neither helps once a message is decrypted and displayed. At that point, anything that can read the screen can read the message.
Whittaker’s post names the features Signal is reacting to: Google’s Gemini Intelligence and Apple’s Onscreen Awareness for Siri AI. These assistants “scan” what is on screen so they can answer questions and take actions. Signal’s concern is what happens next. In Whittaker’s words, the data is “captured, stored, and made accessible to whatever local database or remote server an OS-level AI agent is using at that moment,” entirely outside Signal’s control. That affects not only the user who invoked the assistant but everyone else in the conversation, who never agreed to anything.
Signal is also worried about third-party agents. The post points to agents such as Meta’s Muse that ask users to grant deep permissions during setup, often through accessibility settings. Signal argues those settings were built to help people with disabilities use apps and are now being used as a “skeleton key” for AI access.
Signal has done this before
This is the second time Signal has turned to screenshot blocking. In May 2025 it enabled a Screen security setting by default in Signal Desktop on Windows 11 to keep Microsoft Recall from capturing chats. Recall takes frequent snapshots of the screen, extracts the text and stores it in a searchable database. Signal says Recall could retain Signal content indefinitely, including disappearing messages. At the time, it explained that it had to borrow a digital rights management flag normally used to protect streamed video because Windows offered no better tool.
The new move extends that logic from one desktop feature to the two dominant mobile platforms. It also highlights a gap. Signal says it cannot reliably block screenshots on macOS: protections against continuous screen capture exist there, but not against ordinary screenshots, which is how most agents read app data, and some older APIs have been deprecated in favor of protections meant for streaming media. For now, Signal’s only advice to macOS 27 users is to add Signal to the Excluded Apps list in the Siri section of System Settings.
Signal admits the remedy is partial
Whittaker was candid that blocking screenshots does not close every door. An agent could, in principle, retrieve content already stored on the device through the file system. Signal’s fix addresses the screen, not every other route into an app’s data.
That is why the post ends with two requests to Apple, Google and Microsoft. First, give developers an easy, documented way to opt an app out of AI data access, covering screen reading, file system access, notifications, APIs and other OS frameworks, without breaking the app. Second, publish proper technical documentation of how these agents access and process data. Signal said it has spent months on “forensic investigations” across Android and iOS just to learn what the assistants do, and it rejected the idea that confidential computing alone solves the problem, because the OS provider still decides how much privacy users get.
There has been some movement on the user side. The Electronic Frontier Foundation reported in October that Apple’s iOS 27.2 update added an “Excluded Apps” list that stops Siri AI from accessing the content of chosen apps. That is a control a user must find and apply, though, not a guarantee an app developer can rely on, which is the distinction Signal keeps drawing. Signal said it is “cautiously optimistic” that Google and Apple are engaging with it.
What this means for compliance and security teams
Most organizations do not ship an encrypted messenger. Many do run phones full of confidential conversations, customer data and regulated records, and the same OS assistants are arriving on those devices. The lesson is broader than Signal. Privacy promises made at the app layer can now be undercut at the OS layer, as other high-profile disputes such as Apple pulling Advanced Data Protection in the UK have already shown.
Practical steps worth taking now:
- Inventory where OS assistants are switched on. Check which managed and personal devices used for work have screen-aware assistants enabled, and whether your mobile device management tools can control them.
- Use the controls that exist. Apple’s Excluded Apps list and the Connected Apps settings in Gemini are imperfect, but they let you keep assistants out of email, messaging and line-of-business apps that handle sensitive data.
- Review your own apps. If you build mobile software that displays health, financial or other sensitive data, test what OS assistants can read and whether screen-capture flags are set where they should be.
- Update acceptable-use and recording policies. Screen-reading agents raise many of the same consent questions as AI transcription tools that record meetings: other participants have not agreed to have their words processed by a third-party model.
- Document AI data flows in vendor and risk assessments. Treat the OS assistant as a processing activity with its own data destinations, retention and training questions, rather than as part of the phone.
It is also worth resisting easy assumptions. “On-device” does not always mean data stays on the device, and an encrypted channel does not protect content once it is displayed. Those are exactly the kinds of beliefs that tend to fall apart, as we noted in our look at privacy myths about AI. Hardware makers are experimenting with physical protections too, such as the privacy display on Samsung’s Galaxy S26 Ultra, but a screen filter does nothing against software that reads the screen from inside.
Finally, the episode fits a wider pattern of pressure on private messaging from different directions. In Europe the debate has centered on proposals that could require scanning of communications, which Norway’s data protection authority warned could turn private messaging into mass surveillance. Signal’s complaint is different in origin, coming from commercial AI design rather than law enforcement, but the effect it describes is similar: content that was meant to be visible only to the people in a conversation becomes readable by someone else.
Frequently asked questions
Does this mean Signal messages were being leaked to Apple or Google?
Signal has not said that. Its concern is that OS assistants can capture on-screen content and send it to local databases or remote servers beyond Signal’s control. The block is preventive.
Can I still take screenshots in Signal?
Yes. On iPhone you can turn Block Siri AI off entirely or for 10 minutes. On Android you can turn Screen security off in settings. Signal asks you to confirm and warns that assistants may then be able to read your chats.
Why doesn’t the protection work on a Mac?
Signal says macOS lacks a clearly documented way for apps to block ordinary screenshots. It suggests Mac users add Signal to the Excluded Apps list in the Siri settings instead.
Should companies block OS assistants on work phones?
That is a risk decision rather than a rule. Organizations handling confidential or regulated data should at least know where assistants are enabled, exclude sensitive apps where possible and record the choice in their AI and device policies.
The canary speaks up
Whittaker called Signal “the canary in the coal mine,” and the label fits. A nonprofit messenger with a narrow mission is the first to say publicly that OS-level AI has changed the rules for every app that handles private data. Signal can block a screenshot. Whether app developers get a reliable off switch for AI access will be decided by Apple, Google and Microsoft, and it is worth watching closely.