Signal blocks OS-level AI agents from reading private messages on screen
Signal has updated its iOS and Android apps to stop Siri, Gemini and other agents from capturing chats via screenshots, while saying developers still lack robust tools to protect app-level privacy.
· 2 min read
- Signal's iOS and Android apps now block screenshots by default so AI agents such as Siri and Gemini cannot read chats.
- CEO Meredith Whittaker says the fix is only partial and that platform makers give developers too few controls.
- Apple is adding tighter permission prompts for data access in macOS 27.3, currently in public beta.
Signal, the encrypted messaging app, has changed its mobile apps so that AI agents built into phone operating systems cannot read messages unless the user explicitly allows it. The Register reported the move on 20 November 2026, saying it covers Apple's Siri, Google's Gemini and other agents.
Signal CEO Meredith Whittaker explained the reasoning in a blog post. She said Google and Apple are building agentic features, including Gemini Intelligence and Onscreen Awareness for Siri, deep into their operating systems. To answer questions and carry out tasks, these agents scan what is shown on the screen, which means Signal content can be captured and handed to them.
The concern, according to the report, is that private messages taken in this way may end up on someone else's server, where they are no longer private. The updated iOS and Android apps now block screenshots by default, so Siri on iOS and Gemini and other agents on Android cannot take them. Such agents typically process screenshots with optical character recognition to pull out the text.
Users who want to capture their own conversations can switch the protection off in Signal's settings. Whittaker warned that doing so could expose both their own messages and those sent by their contacts to AI agents. Signal previously introduced similar screenshot blocking to guard against Microsoft Recall.
Whittaker acknowledged the approach is only a partial fix, since agents may find other routes to application data, such as the device file system. She called it a serious problem that Signal and other developers lack robust tools to stop AI agents from undermining app-level privacy. She said Google, Apple, Microsoft and others offer no easy way to restrict such access without drastic compromises to the core user experience, and that this is why Signal cannot yet apply the protections on macOS.
Apple is taking some steps on macOS, though The Register says they appear aimed mainly at third-party access rather than at limiting Siri, which can be turned off separately. In macOS 27.3, now in public beta, the Device Control and Data Access and Full Disk Access APIs will get more granular controls. Users will see more permission prompts, and developers of backup software and accessibility tools will need to apply to Apple for an entitlement to obtain Full Disk Access.
Apple said that as AI agents grow more capable and autonomous, the risks of such access will increase substantially, and that it wants users to understand those risks before granting it. The Register also noted reports that Apple is building its own AI agent to compete with Anthropic and OpenAI.
The report places Signal's decision within a wider pushback against agents. Amazon has blocked shopping agents from its e-commerce site, while Walmart has invited third-party agents in through its AI Gateway. United Airlines has banned transactions from unauthorized AI bots, and other retailers and financial firms have set their own limits. Bot detection vendors such as Cloudflare, Akamai and DataDome are also seeing growing demand.