Signal blocks AI agents from accessing private messages

A cracked smartphone screen with a translucent hand reaching through from the other side, trying to grasp encrypted messages that glow faintly behind the glass; dim ambient light emphasizes tension be

Signal is taking technical measures to block AI agents built into mobile operating systems from accessing user messages, citing an existential threat to privacy. The app is now preventing screenshots on iOS 27 and restricting screen access when AI assistants attempt to read content on Android. This follows similar action last year against Microsoft’s Recall feature on Windows 11, which continuously captured screen images and stored them in a searchable database.

AI integration undermines app-level privacy

Google and Apple are embedding agentic AI systems—such as Gemini Intelligence on Android and Onscreen Awareness with Siri AI on iOS—deep into their operating systems. These systems scan on-screen content to respond to user requests, effectively granting them access to everything visible in apps like Signal. Because Signal relies on end-to-end encryption and open-source verification to guarantee privacy, this unauthorised screen access bypasses all built-in protections. Once data is captured by the OS-level AI, it can be stored locally or transmitted remotely, entirely outside Signal’s control.

This shift transforms the operating system from a neutral platform into an active data collector. For Signal, which operates as a nonprofit under the Signal Technology Foundation, this represents a fundamental challenge: the very environment in which it functions is now hostile to its core mission.

Technical workarounds with limitations

To counter this, Signal has activated its “Screen security” feature on Android when AI-driven screen access is detected. On iOS 27, the app blocks screenshots by default. Users can disable this protection in the app’s settings, but doing so risks exposing their messages—and those of their contacts—to AI systems. These measures are partial: blocking screenshots does not prevent other data access methods, such as reading cached content from the device’s filesystem.

Signal cannot implement equivalent protections on macOS. Apple does not provide a reliable API to block screenshots outside of DRM-protected media, and has deprecated earlier tools. The only available mitigation for macOS 27 users is to manually add Signal to the Excluded Apps list in System Settings > Siri to prevent proactive access by Siri AI—a step most users are unlikely to take.

What it means

For businesses and users in the UK, UAE, US, and Kazakhstan relying on secure communication, this development signals growing tension between AI convenience and fundamental privacy. Signal’s actions highlight that even the most trusted encrypted apps cannot fully protect data if the underlying OS undermines them. Organisations prioritising data sovereignty should assess whether their mobile ecosystems allow such unchecked AI access.

Developers face increasing pressure to adapt without adequate tools. While Google and Apple are reportedly engaging with Signal on improving AI-agent interactions, no formal commitments have been made. Until standardised, transparent controls exist, privacy-sensitive applications remain vulnerable.

Signal’s struggle is a warning: when operating systems become data-harvesting platforms by default, no app-level encryption can fully compensate. The lack of developer control sets a dangerous precedent across the software ecosystem. Independent developers need opt-out mechanisms and clear documentation to uphold user privacy. Until then, users must actively manage settings—especially on macOS—to protect their data.

Sources

Leave a Reply

Your email address will not be published. Required fields are marked *