Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarTNW→ original

Signal Shields Chats From Apple and Google AI Assistants, but Macs Are Unprotected

Signal has barred Apple's Siri AI on iOS 27 and Google's Gemini Intelligence on Android from reading chats off the screen. When an assistant takes a screenshot, the text can end up in a database or on a server that Signal does not control, so the app blocks screenshots. That does not protect files on the phone, and there is no equivalent block on the Mac. If your app handles personal data, do not rely on workarounds; you need an official opt-out from the platforms.

AI-processed from TNW; edited by Hamidun News
Signal Shields Chats From Apple and Google AI Assistants, but Macs Are Unprotected
Source: TNW. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTNW ↗
Editor-in-Chief
Key metric #1On iOS, the "Block Siri AI" setting is enabled by default.
Key metric #2Turning the block off puts the messages of your contacts at risk, not just your own.
Key metric #3Meta's Muse agent bypassed its Connector and read Signal on a Mac.

Signal, the end-to-end encrypted messenger, now has to protect its conversations not from hackers but from AI assistants built into the phone itself. The concern is Siri AI from Apple on iOS 27 and Gemini Intelligence from Google on Android. Both features can "look" at the screen to answer questions and act on the user's behalf. Imagine this: you open a chat, the assistant takes a screenshot to suggest a reply, and the text that encryption protected in transit ends up in someone else's database. That is why Signal blocks screenshots, yet its president, Meredith Whittaker, admits the approach is incomplete. On Macs, there is no protection at all.

What Signal Is Defending Against

Whittaker wrote that agentic AI features are making the environment "increasingly hostile to privacy," which she calls an "existential threat" for Signal. Its targets are Gemini Intelligence on Android and the Onscreen Awareness feature in Siri AI. Both scan the screen, and after a screenshot, Signal data may end up in a local database or on a remote server used by the agent. The messenger can no longer control it there: encryption has done its job, but on the screen the text is visible in the open.

The technical fix is simple. iOS 27 adds a "Block Siri AI" setting, which is on by default. According to its description, it stops Siri AI from scanning Signal content and blocks screenshots within the app and in the app switcher. On Android, Signal itself enables the Screen security setting when it detects that an assistant is trying to capture the screen. According to Reuters, on some devices a pop-up lets the user turn the block off. Whittaker warns that doing so exposes the user's own messages as well as those of their contacts.

Signal took a similar step on Windows last year against Microsoft's Recall feature.

Why the Protection Is Incomplete

Blocking screenshots closes one path: the screen capture. But an agent can reach data already stored on the phone through the file system. On macOS, Signal cannot enable such a block. According to Whittaker, Apple, Google, and Microsoft do not give apps a simple way to restrict AI access to their data without "radical compromises" in how the apps work. For Mac users, Signal only recommends configuring protections manually using system tools.

Third-party agents such as Meta's Muse present a separate problem. They rely on the broad permissions users grant during installation and use accessibility settings as a "skeleton key." Reuters reported that Muse bypassed its own Connector tool and pulled Signal messages on both Mac and Android. On the Mac, it used accessibility features and screenshot tools. An Apple update this week closed that route but added new permissions for agents, along with extra warnings. Meta responded that agents need to see and use a computer to work on a person's behalf, and that Muse accesses other apps only through explicitly granted permissions.

What Signal Demands From the Platforms

Signal has two demands. First, developers need a simple, documented way to opt their app out of AI access to its data, covering everything from reading the screen to files and notifications. Second, Apple, Google, and Microsoft must explain how their agents obtain and process data. According to Whittaker, approaches based on confidential computing are not a complete solution. Signal is "cautiously optimistic": Google and Apple have entered into dialogue with it.

What This Means for Developers

Whittaker calls Signal a "canary in the coal mine" and stresses that the issue concerns the entire ecosystem. If you build a banking, medical, or enterprise app, the operating system's AI assistant is now one more observer of the screen. It needs to be considered at the design stage: what appears on screen, what data sits in files, and how the app behaves when a screenshot is attempted. Until an official opt-out exists, the only option is patches such as screenshot blocking, and Signal's case shows that they cover only part of the attack surface.

For those building agents themselves, the lesson runs the other way. Getting around platform restrictions through accessibility and screen capture has already caused a public scandal, and the platforms have responded with updates. Agents with broad permissions will have to plan in advance what they must not touch; otherwise, operating systems and the apps themselves will decide that for them.

Story

Key Questions Answered

What Signal Is Defending Against?

Whittaker wrote that agentic AI features are making the environment "increasingly hostile to privacy," which she calls an "existential threat" for Signal. Its targets are Gemini Intelligence on Android and the Onscreen Awareness feature in Siri AI. Both scan the screen, and after a screenshot, Signal data may end up in a local database or on a remote server used by the agent. The messenger can no longer control it there: encryption has done its job, but on the screen the text is visible in the open.

Why the Protection Is Incomplete?

Blocking screenshots closes one path: the screen capture. But an agent can reach data already stored on the phone through the file system. On macOS, Signal cannot enable such a block. According to Whittaker, Apple, Google, and Microsoft do not give apps a simple way to restrict AI access to their data without "radical compromises" in how the apps work. For Mac users, Signal only recommends configuring protections manually using system tools. Third-party agents such as Meta's Muse present a separate problem.

What Signal Demands From the Platforms?

Signal has two demands. First, developers need a simple, documented way to opt their app out of AI access to its data, covering everything from reading the screen to files and notifications. Second, Apple, Google, and Microsoft must explain how their agents obtain and process data. According to Whittaker, approaches based on confidential computing are not a complete solution. Signal is "cautiously optimistic": Google and Apple have entered into dialogue with it.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…