KEY POINTS
- Signal's iOS and Android apps now block AI agents from taking screenshots of chats by default.
- CEO Meredith Whittaker calls the fix partial because operating system makers give developers few ways to restrict AI access.
- Apple is adding tighter permission controls in macOS 27.3, which is in public beta.
Signal, the encrypted messaging app, has changed its iOS and Android apps so that AI agents built into the operating system cannot read messages unless the user explicitly allows it. The Register reported that the move targets Google Gemini, Apple's Siri and similar agentic tools, and is motivated by privacy concerns.
Signal CEO Meredith Whittaker compared the step to the screenshot blocking the company introduced last year to defend against Microsoft Recall. In a blog post, she said Google and Apple are building agentic features such as Gemini Intelligence and Onscreen Awareness with Siri deep into their systems. These agents work by scanning what is shown on a screen, which means Signal content can be captured and passed to them.
According to Whittaker, the consequence is that private messages processed by these systems may end up on someone else's server, where they are no longer private. The updated apps now block Siri on iOS and Gemini and other agents on Android from taking screenshots, which the agents process with optical character recognition to pull out text.
Users who want to screenshot 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.
She also acknowledged that blocking screenshots is only a partial answer, because agents might reach application data through the device file system. She described it as a serious problem that Signal and other developers lack strong tools to stop AI agents from undermining app-level privacy. According to her, Google, Apple, Microsoft and others do not let developers restrict such access without drastic compromises to the core user experience, which is also why Signal cannot yet offer the protection on macOS.
Apple is taking some action on macOS, The Register noted, though mainly to limit third-party access to data rather than to restrict Siri, which can be turned off separately. With macOS 27.3, now in public beta, the system will add more granular controls for Device Control and Data Access and for Full Disk Access. Users will see more permission prompts, and developers of backup software and assistive tools that rely on Accessibility APIs will need to request an entitlement from Apple.
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 them before granting permission. The Register also said Apple is reportedly developing its own AI agent to compete with Anthropic and OpenAI.
The outlet placed the Signal change in a wider pattern of resistance to third-party agents. Amazon has blocked shopping agents from its e-commerce site, while Walmart has invited outside agents in through its AI Gateway. United Airlines has banned transactions from unauthorized bots, and Target, Home Depot, Etsy, eBay and several large banks have set up their own limits. A bot detection industry led by companies such as Cloudflare, Akamai and DataDome is also growing.