Signal locks screenshots to counter AI agents
Signal blocks screenshots on iPhone and Android to limit AI assistants' access to private conversations. The publisher considers the threat serious but acknowledges incomplete protection.
Signal now blocks screenshots in its app on iOS 27 and, on Android, when it detects an AI assistant trying to read the screen. The publisher aims to prevent Siri AI and Gemini from accessing private messages, while system-integrated agents are becoming, according to Meredith Whittaker, « increasingly hostile to privacy ». The issue goes far beyond Signal: it concerns how Apple, Google, and Microsoft allow their assistants to probe data displayed on the screen.
Le nouvel écran de fumée de Signal vise les assistants, pas les curieux
On iPhone, Signal enables by default a “Block Siri AI” setting in iOS 27. This option prevents Siri AI from scanning the app’s content, also blocks screenshots, and hides Signal in the app switcher. On Android, the app relies on its Screen security feature, which triggers when it detects an assistant attempting to read the screen. Users can disable the protection, but Signal warns they then expose their messages, and those of their contacts, to these agents.
The logic is simple: if the assistant can see what is displayed, it can also ingest conversations meant to remain private. Meredith Whittaker, Signal’s president, speaks of an « existential threat to Signal » in a post published Friday. She mainly targets system integrations like Gemini Intelligence on Android and Siri’s Onscreen Awareness feature, which read what is displayed to respond to the user or act on their behalf. Once captured, this data can end up in a local database or on a remote server controlled by the agent, outside Signal’s scope.
Une protection utile, mais loin d’être étanche
Signal admits frankly: its countermeasure remains incomplete. Blocking a screenshot does not prevent an agent from accessing data already stored on the phone via the file system. On some Android devices, the app also cannot reliably detect all access attempts. The security veneer thus holds, but it cracks at the edges.
The publisher had already taken a similar measure on Windows last year against Microsoft Recall, the system that regularly records screenshots to reconstruct a searchable activity timeline. However, Signal cannot apply the same defense on macOS. The app then recommends adding Signal to the list of apps excluded in Siri settings. According to Reuters, Apple has already closed part of the access via accessibility tools, following reports notably targeting Meta’s Muse agent, which could have retrieved messages on Mac and Android via accessibility and screenshot permissions.
Signal met Apple, Google et Microsoft face à une demande très concrète
At its core, Signal demands two things: a clear, documented method to exclude an app from AI agents’ access, and more transparency on how these agents collect and process data. Whittaker believes that confidential computing approaches — mechanisms meant to process data in protected environments — are insufficient. She also wants tools to limit third-party agents, including those relying on permissions granted at installation. Accessibility permissions are becoming, in her view, a « skeleton key ». The master key, in short.
Signal says it remains « cautiously optimistic » about discussions with Google and Apple. Yet the demand is politically sensitive: if system publishers must allow apps to opt out of AI agents, they will also have to explain the extent of their own assistants. In this matter, encrypted messaging serves as a real-world test. Signal describes itself as « the canary in the coal mine », but the issue far exceeds its case.
By the numbers
- iOS 27 — version where Signal adds “Block Siri AI”.
- 2025 — year when Signal blocked Microsoft Recall on Windows.
- 2 platforms — iOS and Android, affected by the new protection.
- 3 giants — Apple, Google, and Microsoft, challenged by Signal.