Signal Blocks Screenshots to Keep AI Agents From Reading Your Chats

Signal is blocking screenshot capture in its Android and iOS apps so that AI agents cannot read users' messages and other data, according to The Verge, which published the report on November 20, 2026.
The target is on-screen AI. Signal's screenshot blocking on iOS 27 and on Android is meant to stop Siri, Gemini, and other Android AI assistants from reading what is displayed in the app. The change is a control applied inside Signal's own client, not a change to the operating systems that host it.
Signal president Meredith Whittaker said that agentic AI capabilities which analyze on-screen information are undermining Signal's ability to independently guarantee privacy. The company's position, as reported, is that end-to-end encryption, which scrambles a message so only the sender and recipient can read it, protects the message in transit and at rest. It cannot control what a separate process reads from the screen once the message has been decrypted for display.
Users can restore screenshot capability in Signal's app settings. Whittaker warned that doing so will potentially expose the user's Signal content, and that of their contacts, to AI agents. The setting moves the privacy decision from the platform to the individual user, and the cost of re-enabling it falls on people who did not write the messages in question.
Whittaker also acknowledged that the remedy is incomplete. Blocking screenshots does not address other routes AI agents can use, such as retrieving data through the device filesystem. The restriction covers one retrieval path among several.
Signal said it cannot yet implement these AI-agent protections on macOS, so the desktop client lacks the same control. That is a gap for users who run Signal across several devices.
This is not the first time Signal has acted against a platform-level capture feature. The company previously blocked Microsoft's Recall, which periodically records screen content for later search, from capturing secured chats. Recall raised the same underlying question: who, or what, is entitled to read a decrypted message on a device?
Signal is calling on Apple, Google, and Microsoft to provide simple technical tools and accurate documentation that would help app developers restrict AI data access. As reported, the request is directed at the platform vendors. Signal has not described a specific API or policy it expects them to adopt.
The broader context here is a shift in the threat model. End-to-end encryption has long assumed that the endpoints are trustworthy and that the main adversaries are network observers and server operators. An on-device assistant with screen-reading permission adds a reader that the user has authorized, sitting inside the trust boundary, and app developers have few technical levers over it. Signal is narrowing what that reader can capture while asking the operating system vendors for the hooks that would make the restriction durable.
In my view, the most useful question for developers is which permissions their apps can actually enforce. A screenshot flag is a blunt tool, and Whittaker's admission that filesystem access remains open makes that plain. Platform vendors are the parties able to make per-app data boundaries enforceable for AI features, and their answers will shape whether privacy-focused apps can offer real guarantees on the devices people carry every day. Assistants that act on what is on screen can do useful work, such as summarizing a long thread or handling a scheduling request. The open question is whether users can get that help without giving up control of conversations that involve other people.


