SIAgentNews.com
← All news

Signal blocks OS-level AI agents from reading private messages

The encrypted messaging app is implementing new blocks to prevent Siri and Gemini from scanning your screen and capturing chat data.

· First reported by gHacks

iOS 27Version where screenshots are blocked

The short version

  • Signal now blocks screenshots by default on iOS and prevents AI screen capture on Android.
  • The move targets 'agentic' features like Gemini Intelligence and Siri's Onscreen Awareness.
  • Signal warns that OS agents can move captured data to local databases or remote servers.
  • The company is calling on Apple, Google, and Microsoft to provide tools to restrict AI data access.
  • Users can opt-out of these protections in the app settings but may expose messages to AI agents.

Why it matters

This change highlights a growing conflict between convenient AI features and end-to-end encryption. It signals a shift where app developers must actively fight the operating system to maintain user privacy.

The rise of agentic OS features

Signal has launched a series of updates to prevent AI agents built into mobile operating systems from reading user messages. The company is responding to a trend where tech giants are weaving 'agentic' capabilities deeper into the core of their software. These features are designed to help users complete tasks by understanding what is happening on their screens at any given moment.

Specifically, Signal is targeting tools like Google Gemini Intelligence and Apple's Onscreen Awareness for Siri. These systems are designed to scan the information currently displayed on a device to provide answers or perform actions. When a user has a chat window open, these agents can see the text and images within that conversation as they appear on the display.

The company describes these integrations as an existential threat to its privacy model. While Signal uses encryption to protect data while it is moving between devices, it cannot prevent an operating system from seeing what is shown on the screen. This creates a loophole where the privacy of all participants in a conversation could be compromised by the device itself.

How the new blocks work

To combat these risks, Signal is implementing specific technical barriers depending on the mobile platform. On iOS 27, the app now blocks screenshots by default. This prevents the system from capturing a still image of a private conversation to be stored or processed by other services.

The approach for Android is slightly different and focuses on active detection. Signal now blocks screen capture specifically when it detects an AI assistant attempting to read the screen. This allows the app to remain functional for standard use while creating a wall against automated AI data collection.

Users who still want to take screenshots can manually turn these protections off within the Signal settings. However, the company warns that doing so can expose both the user's and their contacts' messages to AI agents. This creates a clear choice between the convenience of screen capture and the security of the app's privacy guarantees.

Data flow and privacy risks

The primary concern for Signal is what happens to the data once it is captured by an OS-level agent. Once an AI assistant scans a message, that information is handed over to the agent's processing pipeline. From that point forward, the data is outside of Signal's control and could end up in a local database or on a remote server.

Signal notes that this is a serious problem because developers currently lack robust tools to prevent AI agents from undermining application-level privacy. The company argues that Google, Apple, and Microsoft do not provide an easy way for developers to restrict this access. Without these tools, apps are forced to make drastic compromises to the core user experience just to keep data private.

The company has already taken similar actions on other platforms to address these concerns. Last year, Signal turned on screenshot blocking by default on Windows 11. This was specifically intended to stop Microsoft's Recall feature from taking and storing screenshots of messages on the desktop.

Limitations and next steps

Signal admits that the current fix is incomplete and does not solve every potential vulnerability. For instance, blocking screenshots does not stop an AI agent from finding other ways to reach app data. An agent might still be able to read content stored on the device through the file system if it has the necessary permissions.

The company is currently calling on the three major tech giants to provide simple technical tools and accurate documentation. They want these tools to allow developers to keep AI agents away from sensitive app data without breaking the app's functionality. They are also seeking protection against third-party agents, such as Meta's Muse, which use accessibility permissions to read other apps.

Signal is currently in talks with Google and Apple regarding how these systems should interact with privacy-focused applications. The company hopes to share more information about these discussions as they progress. For now, the focus remains on creating a technical environment where users can enjoy AI assistance without sacrificing the privacy of their private conversations.

What to watch

  • Whether Apple and Google release official APIs for developers to block AI screen scanning.
  • How other encrypted apps like WhatsApp or Telegram respond to these OS-level AI features.
  • The development of screenshot protection for the macOS version of Signal.

Try it

Check your Signal settings to see if screenshot blocking is enabled and decide if you want to opt-out.

Sources and links

Read the original at gHacks

First reported by gHacks. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news