Signal says it will restrict screenshots by default so that AI agents built into phones cannot read private chats, and it admits the fix is "incomplete." The announcement came on November 20, 2026, in a Signal blog post by president Meredith Whittaker titled "Signal is Serious About Protecting Your Privacy." The change targets OS-level assistants such as Apple's Siri AI and Google's Gemini Intelligence, plus third-party agents that lean on accessibility permissions.
For readers who use Signal, the short version is below. For builders, the story is bigger: an app with a strong privacy promise is telling the operating system vendors that it cannot reliably keep agents out, and it wants a documented opt-out.
Quick facts #
| Question | Answer |
|---|---|
| What changed? | Signal restricts screenshots by default on phones to stop AI agents capturing chats |
| iOS | iOS 27: blocks Siri AI by preventing screenshots in the app; a "Block Siri AI" toggle in Settings |
| Android | Detects an AI assistant reading the screen, then enables "Screen security" automatically |
| macOS | Not covered; Signal cannot reliably block screenshots there |
| Windows | Already blocks Microsoft Recall, on by default since last year |
| Can I turn it off? | Yes, but Signal warns it exposes you and your contacts |
| Signal's own verdict | "Incomplete": files on the device can still be reached |
| What does Signal want? | Easy, documented opt-outs from Apple, Google and Microsoft |
What did Signal change on each platform? #
According to Signal's blog post, the response differs by operating system, because each one exposes different hooks.
iOS. On iOS 27, Signal blocks Siri AI from capturing its content by preventing screenshots inside the app. Settings includes a "Block Siri AI" toggle. The post shows the toggle enabled but does not say whether it ships on by default. We covered Apple's wider push in Siri model delegation to Claude and ChatGPT.
Android. Signal detects when an AI assistant tries to read the screen and then switches on its "Screen security" setting automatically. The company explains why it did not simply block screenshots everywhere: on many vendor builds outside Pixel it cannot reliably detect screenshot attempts once screenshots are already disabled, and blocking them outright would confuse users. So coverage is not universal across Android devices. The named feature is Gemini Intelligence with Google Gemini, which we explained in our post on Android 17 and Gemini Intelligence.
macOS. Signal does not block Siri AI on a Mac. It says Apple's available screen-protection APIs cover streaming and DRM content, not screenshots. The workaround is to add Signal to the Excluded Apps list under System Settings, then Siri.
Windows. Signal already used screenshot blocking last year to stop Microsoft Recall, which takes screenshots every few seconds and runs text recognition to build a searchable database. That protection is on by default and could otherwise have retained Signal content, including disappearing messages.
Why does Signal call its own fix incomplete? #
The company is blunt. Blocking screenshots does not stop other access routes, such as an agent retrieving content already stored on the device through the filesystem. Signal says it lacks robust tools to stop AI agents from undermining app-level privacy without serious compromises to the user experience. It also says the agents are poorly documented, so its developers used forensic analysis on Android and iOS to learn what they do.
That honesty matters. A screenshot block is a patch on one channel, while an agent with accessibility permissions, notification access or file access has several. Signal's wider argument is the one Whittaker has made for a year: agents need deep access to work, and that access breaks the boundary between apps and the operating system. We tracked the earlier version of that argument in our guide to indirect prompt injection and agent risks.
How we got here: a year of warnings from Signal #
This is not a sudden turn. In May 2025 Signal published By Default, Signal Doesn't Recall, explaining how it used Windows' DRM flag to stop Microsoft Recall from capturing its window. Signal's argument then was that Microsoft gave app developers no other option.
Whittaker then widened the case from one feature to a whole class of software. In November 2025 Fortune reported her saying that AI agents are an existential threat to secure messaging. The point of that argument: an agent that acts for you has to read your screen, your contacts and your messages, and that dissolves the wall between apps. The November 20 change is the first time Signal has turned that argument into a default-on screenshot restriction on phones.
The Meta Muse context #
Signal's post does not say why it acted now, and we did not find a source that does. The timing follows Reuters reporting on Meta's Muse agent, which we summarized in Meta Muse reading Signal messages. Treat the link as context, not a confirmed cause.
Signal's blog post says it also wants tools to stop third-party agents from using accessibility permissions to read app data. That is the same class of access that agent tools use for computer control, as in Vercel's accessibility-tree desktop agent.
Agents that act on your behalf need rules. See how explainx helps secure your agents →
What does Signal ask Apple, Google and Microsoft to do? #
Signal lists two requests of the operating-system vendors:
- A simple, documented opt-out. Apps should be able to opt out of AI data access and processing, including screen reading, filesystem access, notifications, APIs and other frameworks.
- Better documentation. Vendors should publish technical specifications of how agents access and process data. Signal argues that confidential computing does not fully solve the concern.
The company says it is cautiously optimistic that Google and Apple are engaging with it, and it credits allies including the EFF for pushing similar changes. Neither vendor's response is in the sources we reviewed, so we do not report one.
Why this matters for builders and teams #
If you ship an app that shows private data, a user's agent may now see it. Consider what that means:
- Threat model. Treat "the OS assistant can read my screen" as a default capability, not an exotic attack. A prompt-injection or a leak in that assistant becomes a leak of your app's content.
- Platform gaps. There is no cross-platform API to say "do not read this window." Signal had to rely on screenshot flags and detection tricks that differ per OS.
- Enterprise exposure. Work chats on a phone with an always-on assistant may be processed outside your control. Our reporting onGrok bot permissions in Slack shows how quickly agent access turns into a data question.
How to protect yourself: a short checklist #
- iPhone: update to iOS 27, open Signal Settings and confirm Block Siri AI is on.
- Android: make sure Screen security is enabled, and update Signal. Note that coverage varies by device.
- Mac: add Signal to the Excluded Apps list in System Settings, Siri.
- Windows: leave Signal's screen security on if you run Recall.
- Review accessibility and notification permissions. Remove them from any third-party agent that does not need them.
- Use disappearing messages and avoid forwarding sensitive text to chat with an assistant.
- For teams running agents, a runtime guard helps: AgentBeam covered the related permission path inhow Meta Muse reads Signal through notifications and OS permissions , and it is the agent security platform from the explainx.ai team that stops AI agents before they take dangerous actions.
What people are asking #
Is my Signal chat still end-to-end encrypted? Yes. Encryption protects messages in transit. The risk here is an assistant reading the decrypted text on your screen or device, which encryption does not address.
Will turning on screen security break my screenshots? On Signal, yes: you may not be able to screenshot chats. That is the trade-off Signal accepts by default, and you can disable it.
Does this stop Meta Muse? Only partly. Signal says screenshot blocking does not cover all routes, and it wants tools to stop third-party agents from using accessibility permissions.
The bigger picture #
This is the second time in two years that Signal has used a blunt product feature against an OS-level AI. The Windows move against Recall showed it works for screenshots. The 2026 move shows the limit: once agents combine screen reading, accessibility, notifications and files, an app alone cannot win. That is why Signal is pushing for platform rules rather than just shipping toggles.
Expect other privacy-first apps to follow with their own toggles, and expect the OS vendors to be asked, again, why an app cannot say no. We will add a dated update if Apple, Google or Microsoft publish a developer opt-out.
Related reading #
Details are as of November 20, 2026, and platform behavior may change with updates.