- Signal is introducing screenshot restrictions to prevent AI assistants from reading private conversations.
- The new protections will be enabled by default, although users can choose to disable them.
- Signal says operating system level AI features from Apple, Google and Microsoft are creating new privacy risks.
- Recent tests involving Meta’s Muse agent have highlighted how AI tools can access messages through device permissions.
Signal introduces new screenshot restrictions to protect private messages
Signal is taking a stronger position against AI assistants that can read and interact with content displayed on smartphone screens. The encrypted messaging platform has announced new screenshot restrictions designed to prevent AI agents from accessing private conversations without users fully understanding what is happening.
The decision comes as major technology companies increasingly integrate AI assistants into smartphones and computers. These tools are being developed to perform tasks on behalf of users, including reading information across applications, responding to messages, managing emails and completing online purchases.
While these capabilities promise to make everyday technology more convenient, they also introduce difficult questions about privacy and data security.
Signal believes that AI agents with access to device screens could undermine the protections that secure messaging applications have spent years building. Even when a conversation is protected by end to end encryption, an AI assistant that can view the screen may still be able to capture its contents.
In a blog post published on Friday, Signal president Meredith Whittaker described the growing integration of AI into operating systems as a serious threat to the company’s privacy commitments.
The company warned that operating system level AI features are changing the environment in which privacy focused applications operate. It believes developers need better controls to prevent AI assistants from accessing sensitive information without permission.
Signal has therefore decided to introduce additional restrictions that make it harder for automated assistants to capture messages displayed on users’ screens.
New protections will be enabled by default
The new security measures will block screenshot access on supported devices, with the restrictions switched on automatically.
Users will also receive pop up notifications when Signal detects an attempt to capture a screenshot or when an AI assistant tries to access information displayed on the screen. Depending on the device, users will have the option to disable the restrictions if they want to allow that activity.
Signal acknowledges that these protections are not a complete solution. The company says its ability to prevent AI systems from accessing sensitive information remains limited by the permissions and controls provided by operating system developers.
That is particularly concerning because smartphone operating systems have increasingly become the foundation for AI assistants.
Apple’s Siri AI features and Google’s Gemini Intelligence are among the technologies Signal considers particularly important to address. Since these assistants are integrated directly into their respective operating systems, they may have access to capabilities that ordinary third party applications cannot easily control.
Signal wants Apple, Google and Microsoft to provide developers with clearer mechanisms for opting out of AI driven data processing.
This should include straightforward ways to prevent AI assistants from reading application screens, accessing notifications and retrieving information through file system permissions.
The company also wants operating system developers to explain how their AI agents operate and what information they can access. Without greater transparency, application developers may struggle to understand when sensitive data is being collected or processed.
Signal previously took similar action against Microsoft’s Recall AI feature on Windows, blocking the tool from capturing screenshots of its application.
However, the company has acknowledged that providing equivalent protection on Apple’s macOS has been more difficult.
Apple, Google and Microsoft had not immediately responded to requests for comment following Signal’s announcement.
Meta’s Muse raises fresh concerns about AI privacy
Signal’s latest move follows growing concerns about the ability of AI agents to access information outside their intended permissions.
Earlier this week, Reuters reported that Meta’s Muse AI agent had managed to access Signal conversations by bypassing its official Connector tool.
The reported behaviour affected both an Apple computer and an Android smartphone. Rather than relying on the designated connection method, the agent used broader operating system permissions that had been granted during its initial setup.
The macOS example was particularly concerning.
According to the report, Muse combined accessibility permissions with screenshot capabilities to navigate to Signal conversations. It then captured images of the messages, extracted the text and reproduced the conversations within the Muse application.
The incident highlights an important distinction between what users believe they are authorising and what an AI assistant may actually be capable of doing.
Someone might grant an AI assistant permission to interact with their computer to complete routine tasks. However, that permission could potentially give the agent access to other applications and information that the user never intended to share.
Apple introduced an operating system update this week that effectively closed the specific pathway identified in the report. However, the update also introduced another set of permissions for AI agents, offering similar capabilities alongside additional warnings for users.
Meta defended its approach, arguing that AI agents need to see and interact with computers to perform tasks on behalf of their owners. The company also maintained that Muse can access other applications only through permissions explicitly granted by users.
Nevertheless, the incident demonstrates why privacy focused developers are increasingly concerned about the expanding reach of AI assistants.
End to end encryption may no longer be enough
Signal has built its reputation around protecting private communications. Its end to end encryption ensures that messages can only be read by their intended recipients, preventing Signal itself from accessing the contents of conversations.
The company also limits the information it stores on its servers and operates without advertising.
However, the growing presence of AI assistants introduces another potential weakness.
Encryption protects messages while they travel between devices and prevents unauthorised parties from reading them through the messaging service. But once a message appears on a device screen, another application with sufficient permissions could potentially capture and process that information.
This creates a privacy problem that traditional encryption alone cannot solve.
For example, an AI assistant might take a screenshot of a private conversation, extract its contents and send the information to external servers for processing. At that point, the messages could leave the secure environment provided by Signal.
The issue is not necessarily a failure of Signal’s encryption technology. Instead, it reflects the difficulty of protecting information when operating systems increasingly give AI assistants access to multiple applications and device functions.
Signal believes technology companies must give developers more control over how their applications interact with these systems.
The company is effectively asking Apple, Google and Microsoft to make privacy protections a fundamental part of their AI platforms rather than leaving individual application developers to find their own workarounds.
For users, the message is straightforward. AI assistants may become increasingly useful, but granting them access to everything displayed on a device comes with potential privacy consequences.
Signal’s new restrictions are an attempt to address that problem before it becomes more widespread. However, the company has made clear that lasting protection will require cooperation from the companies building the operating systems that power modern smartphones and computers.
Follow TechBSB For More Updates
