Signal is implementing technical restrictions to prevent OS-level AI agents in iOS 27 and Android from capturing screen content, extending its prior mitigation against Microsoft Recall on Windows 11. On iOS 27, Signal blocks screenshots by default to stop Apple's "Onscreen Awareness" with Siri AI from reading messages; on Android, it detects AI assistant screen access attempts and automatically enables the "Screen security" setting. Users can disable this protection in settings, but doing so risks exposing Signal content — including disappearing messages — to AI systems that scan screens, extract text via OCR, and populate searchable local or remote databases outside Signal's control. Signal states these agentic integrations (Gemini Intelligence, Siri AI, Recall) undermine its formally verified encryption guarantees by capturing data post-decryption. The nonprofit emphasizes incomplete coverage: macOS 27 lacks reliable screenshot-blocking APIs, forcing reliance on manual Siri exclusion settings, while Android fragmentation limits detection reliability across vendor distributions. Signal calls on Apple, Google, and Microsoft to provide documented opt-out APIs for AI data access (screen reading, filesystem, notifications) and clear technical specifications for agent behavior, warning that accessibility APIs are being repurposed as "skeleton keys" for third-party agents like Meta's Muse. Confidential computing is deemed insufficient as it leaves privacy guarantees under OS vendor control. Signal frames this as an existential threat to application-layer privacy across the ecosystem.

Read original