Signal Blocks AI Agents From Chats: Screenshots Blocked on iOS 27, Protection on Android
Signal is stopping phone AI assistants from reading chats off the screen. On iOS 27, Apple's new system, the messenger blocks screenshots, and on supported Android versions it turns on protection when an assistant tries to capture the screen. Encryption protects messages in transit, but once a message appears on the display it is already decrypted, and Siri AI or Gemini Intelligence can pick up the text. Developers building apps and agents should therefore treat the screen as a separate risk zone.
AI-processed from The Decoder; edited by Hamidun News
Imagine you're chatting in Signal while the AI assistant built into your phone "watches" the screen to help you with your tasks. Encryption is powerless here: a message has to be decrypted for you to read it, and at that moment it can be read by more than just you. That is why Signal has started blocking screenshots on iOS 27 (the new version of Apple's mobile system) and is enabling protection on supported Android versions when an AI assistant tries to access the screen.
What Signal did
The messenger restricts screenshots so that AI agents cannot read private chats. Here, AI agents means operating-system-level assistants such as Apple's Siri AI and Google's Gemini Intelligence. They can capture screen content to carry out a user's tasks: finding an address, an arrangement or an order number in a conversation.
For users this is convenient. But the agent then sees decrypted messages and can process them outside the app. Reuters quoted Signal warning of an "existential threat" to privacy.
Why this undermines encryption
Imagine locking a letter in a safe, then taking it out and laying it on the table in front of a camera. The safe is still secure, but the contents are visible anyway. End-to-end encryption works the same way: the math protects the message in transit, but it does not stop the system from looking at the display. The agent takes the text straight from the screen, bypassing the cryptography altogether.
The risk is not theoretical. According to the source, Meta's AI agent, called Muse, recently captured Signal content on Android and macOS. It used broad permissions granted during setup, including accessibility tools, and on macOS also screenshots. Meta says this happened only with the user's explicit permission.
Here the trust problem becomes visible. Formally, the user agreed, but few people likely realized that granting screen access would also give the agent their private chats. A broad permission granted at first launch becomes access to everything visible on the display.
The protection is incomplete, and Signal admits it
Signal has already blocked Microsoft Recall on Windows, the feature that periodically captures a computer's screen, but it describes its own protection as incomplete. On macOS, screenshots cannot be blocked reliably. Other paths remain open: an AI can reach the file system and notifications, which can also contain message text. One app alone cannot close all of these.
Signal is therefore turning to the platforms. The company argues that Apple, Google and Microsoft should let apps fully shut out AI agents, in a way that is simple and clear to users. In essence, it is asking for an operating-system-level switch that says "don't show this app to AI."
What this means for developers and product teams
First, if you build a messenger, a banking, medical or enterprise app, the threat model has changed. Until now, it was enough to think about traffic interception and device theft. Now a legitimate, system-level AI is watching the screen alongside the user.
Second, those building their own agents will need to be more careful with permissions. If an agent needs screen or accessibility access, request the minimum and explain exactly what it will see. Otherwise you may get the same reaction Meta did: "the user agreed" does not always sound convincing.
Third, platforms will likely face pressure to respond. Until there is a simple standard way to tell the system "this screen is not for AI," each app will invent its own blocks, and those blocks will not work everywhere.
The limits should be noted. The source gives no details on how Signal detects an AI assistant's access on Android or which versions count as supported. Apple's and Google's positions are not reported in the material, and Meta insists it acted with users' consent. But the direction is clear: the line between a helpful AI assistant and a leak of private conversations now runs along the screen, and it will have to be drawn again.
- Nov 21, 2026OpenAI начала выдавать фоновых агентов dots части подписчиков ChatGPT Plus за $20
- Nov 21, 2026Google to Buy 1.99 Billion Emails from Bankrupt Spirit Airlines for AI for $10 Million
- Nov 21, 2026Google Opens Pre-Orders for Gemini AI Glasses: From $395 and Speech Translation in 70+ Languages
- Nov 21, 2026AI Coding Assistant Under $10: Cursor at $8, Xiaomi MiMo at $6 a Month
Key Questions Answered
What Signal did?
The messenger restricts screenshots so that AI agents cannot read private chats. Here, AI agents means operating-system-level assistants such as Apple's Siri AI and Google's Gemini Intelligence. They can capture screen content to carry out a user's tasks: finding an address, an arrangement or an order number in a conversation. For users this is convenient. But the agent then sees decrypted messages and can process them outside the app. Reuters quoted Signal warning of an "existential threat" to privacy.
Why this undermines encryption?
Imagine locking a letter in a safe, then taking it out and laying it on the table in front of a camera. The safe is still secure, but the contents are visible anyway. End-to-end encryption works the same way: the math protects the message in transit, but it does not stop the system from looking at the display. The agent takes the text straight from the screen, bypassing the cryptography altogether. The risk is not theoretical.
What are the key details on The protection is incomplete, and Signal admits it?
Signal has already blocked Microsoft Recall on Windows, the feature that periodically captures a computer's screen, but it describes its own protection as incomplete. On macOS, screenshots cannot be blocked reliably. Other paths remain open: an AI can reach the file system and notifications, which can also contain message text. One app alone cannot close all of these. Signal is therefore turning to the platforms.
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.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.