Signal Just Called AI Agents an Existential Threat. Encryption Does Not Protect Your Screen. Signal published a November 20, 2026 blog post signed by president Meredith Whittaker calling the agentic AI capabilities Apple, Google and Microsoft are building into their operating systems "an existential threat to Signal, and our ability to responsibly uphold our mission." Signal now blocks screenshots inside its iOS 27 app, auto-enables its "Screen security" setting on Android when it detects an AI assistant reading the screen, and said it cannot reliably block screenshots on macOS because Apple provides no clearly documented APIs; it asked the three platform owners to give developers a documented way to opt apps out of AI data access. Signal named Gemini Intelligence on Android and Onscreen Awareness with Siri AI on Apple devices as the systems it is worried about, and said its screenshot blocking "does not address other more sneaky ways AI agents could potentially access application data, such as retrieving content already present on a user's device via the filesystem. My position is simple: encryption protects the message while it travels, not the screen where you read it. An AI agent built into the operating system sees everything you see, and it does not care how strong your cryptography is. For ten years, companies have treated “end-to-end encrypted” as the end of the privacy conversation. This week, the most respected secure messaging app in the world told us it is not. Written with the help of AI tools; ideas, sources and final review by the author. On November 20, 2026, Signal published a blog post signed by its president, Meredith Whittaker, with an unusually blunt message. The agentic AI capabilities that Apple, Google and Microsoft are building deep into their operating systems are, in Signal’s words, “an existential threat to Signal, and our ability to responsibly uphold our mission.” Signal’s answer is to block screenshots inside its own app. Its admission is even more important: that answer is incomplete, and on the Mac it is not possible at all. The facts are worth reading carefully, because they describe exactly how an OS-level agent gets to your data. On iOS 27, Signal now prevents screenshots inside the app. On Android, where many manufacturers ship their own versions of the system, Signal cannot always tell when a screenshot is attempted. So it watches for something more specific: when it detects an AI assistant trying to read the screen, it automatically turns on its “Screen security” setting. According to Reuters, these protections are on by default, and users get a pop-up offering to disable them when Signal detects a screenshot attempt or an AI assistant trying to access the screen, depending on the device. People who want screenshots back can switch the protection off, but Signal warns this will “potentially expose your and your contacts’ Signal content to these AI agents, and break our privacy guarantees.” Signal names the systems it is worried about: Gemini Intelligence on Android and Onscreen Awareness with Siri AI on Apple devices. It explains the mechanism plainly. To learn about you, these agents scan the information on your screen. Your Signal data is captured and then made available to whatever local database or remote server the agent is using at that moment. Once captured, it is outside Signal’s control. Signal already did this last year against Microsoft’s Recall on Windows 11. What is new is the scale: the problem now reaches the phone in your pocket. Most companies announce a fix and declare victory. Signal did the opposite. It wrote that blocking screenshots “does not address other more sneaky ways AI agents could potentially access application data, such as retrieving content already present on a user’s device via the filesystem.” It said it cannot reliably block screenshots on macOS at all, because Apple does not provide clearly documented APIs to do it. The protections that exist on the Mac cover continuous screen capture and DRM-protected media, not the screenshots most agents rely on. For now, Signal’s only advice to macOS 27 users is to add Signal to the Excluded Apps list in System Settings, under Siri. Signal also said it spent months doing “forensic investigations” on Android and iOS just to understand what these AI systems do with app data, because the documentation does not exist. One of the best privacy engineering teams in the world has to reverse-engineer the operating system to know what the built-in assistant reads. Its request to Apple, Google and Microsoft is simple: give developers “an easy and well-documented way to opt their app out of AI data access and processing,” covering screen reading, filesystem access, notifications, APIs and other OS frameworks, without breaking the app. And publish real documentation on how these agents access and process data. Signal’s post also points to third-party agents, naming Meta’s Muse. It warns that these agents use the deep permissions users are encouraged to grant during setup, in particular accessibility settings, which were designed to help people with disabilities, “not to act as a skeleton key.” The concrete examples are already public. Reuters reported earlier this week that Muse was able to access Signal messages on both a Mac and an Android phone by using broad operating system permissions granted at setup. On macOS, it combined accessibility and screenshot tools to navigate to Signal, capture the chats, extract the text and reproduce the messages inside Muse. Reuters adds that an Apple update this week effectively closed that specific pathway. Meta’s response was that agents need to see and interact with people’s computers to work on their behalf, and that Muse can access other apps only via permissions users explicitly grant. TechCrunch, citing another Reuters report, described a Meta employee who got a reminder from Muse to reply to her sister on Signal. “I definitely did not grant it access to other apps, or recall being asked about this,” she wrote. Apple is moving too. With macOS 27.3, it split the old permissions into two separate settings, Accessibility and “System & App Protection Override,” added an indicator when one app controls another, and lets users require permission every time access is needed. According to Bloomberg, enabling third-party computer-use tools for ChatGPT now takes 14 distinct steps, up from four. Some AI companies see that as protection; others see it as friction for competitors. Either way, the lesson for an enterprise is the same: the permission screen has become a security boundary. Replace “Signal” with any application your company runs on a laptop or a phone. Your CRM. Your HR portal. Your board documents. Your banking interface. Every one of these apps has spent years on encryption, access control and audit logs. None of that matters at the moment an agent with screen access reads what is displayed. Here is the uncomfortable part. Most security reviews I see still ask, “Is the data encrypted in transit and at rest?” They rarely ask, “Which software on this device can read the screen, the notifications and the accessibility tree?” Yet those three permissions are now what decides whether an agent sees everything. Accessibility lets software read and control other apps’ interfaces. Notifications put message previews, codes and alerts in front of anything that can read them. Screen capture turns every window into an image that can be converted into text. An assistant with all three is a full-time observer of your employees’ work, whatever the vendor promises about its intentions. I am not arguing against agents. I am arguing that the security model has moved from the network to the screen, and most organizations have not moved with it. Signal is asking the OS vendors for something every company should demand too: a simple, documented switch, per application, that says “no agent reads this.” Not a confidential computing promise, which Signal explicitly says is not a complete solution because it still leaves the provider in control of the privacy guarantees. A switch that the app owner, and the company managing the device, can set and verify. Until that switch exists, enterprises must build the equivalent themselves, with policy, device management and procurement. What I would ask every CIO and CISO to do this quarter: 1. Inventory the agents. For each managed device type, list every AI assistant built into the OS and every third-party agent installed. Include the ones employees installed themselves. 2. Audit three permissions per device. Accessibility, notification access and screen capture or recording. For each, write down which agent has it and who approved it. 3. Classify your sensitive apps. Messaging, HR, finance, legal, customer data, source code. For each, decide whether any agent may read it. The default answer should be no. 4. Use the exclusions that exist today. On Mac, Signal recommends adding Signal to the Siri Excluded Apps list; apply the same logic to your sensitive apps. Find the equivalent controls on every platform you manage and push them through device management. 5. Turn on in-app protections. Where an app offers screen security or screenshot blocking, enable it by default for sensitive roles and explain why to users. 6. Rewrite your vendor questionnaire. Ask every software vendor: does your app support an agent opt-out? Ask every agent vendor: which permissions do you request, what do you store, where, and for how long? 7. Demand the per-app agent switch. Ask Apple, Google and Microsoft through your account teams. Large customers move roadmaps faster than blog posts do. 8. Treat agent permissions like admin rights. Time-limited, approved by a named person, reviewed regularly, removed when unused. Signal called itself “the canary in the coal mine.” That is exactly right, and companies should listen. When the app built entirely around privacy says it can no longer guarantee it on its own, every other app on the same device is in the same situation, usually with less engineering power to respond. Encryption is still essential. But it protects the pipe, not the glass. The next privacy decision in your company will not be made in the data center. It will be made on a permission screen, often in a few seconds, by an employee who just wants the assistant to work. Make that decision for them, deliberately, before an agent makes it by default. If you want more on how to deploy AI agents without losing control of your data, follow me here on Medium. Signal, Meredith Whittaker, “Signal is Serious About Protecting Your Privacy,” November 20, 2026: https://signal.org/blog/protecting-your-privacy/ https://signal.org/blog/protecting-your-privacy/ Reuters via Investing.com , Katie Paul, “Signal restricts screenshots to block AI agents from reading messages,” November 20, 2026: https://www.investing.com/news/stock-market-news/exclusivesignal-restricts-screenshots-to-block-ai-agents-from-reading-messages-5040998 https://www.investing.com/news/stock-market-news/exclusivesignal-restricts-screenshots-to-block-ai-agents-from-reading-messages-5040998 The Verge, Jess Weatherbed, “Signal is blocking Gemini and Siri AI from reading your screen,” November 20, 2026: https://www.theverge.com/ai-artificial-intelligence/1038390/signal-blocking-ai-agent-screenshots-mobile-app https://www.theverge.com/ai-artificial-intelligence/1038390/signal-blocking-ai-agent-screenshots-mobile-app TechCrunch, Aisha Malik, “Meta’s AI agent Muse is mixing up conversations and could be exposing private data, report says,” November 19, 2026: https://techcrunch.com/2026/11/19/metas-ai-agent-muse-is-mixing-up-conversations-and-exposing-private-data-report-says/ https://techcrunch.com/2026/11/19/metas-ai-agent-muse-is-mixing-up-conversations-and-exposing-private-data-report-says/ TechCrunch, Sarah Perez, “Apple reveals new Mac security controls as AI agents raise privacy risks,” November 16, 2026: https://techcrunch.com/2026/11/16/apple-reveals-new-mac-security-controls-as-ai-agents-raise-privacy-risks/ https://techcrunch.com/2026/11/16/apple-reveals-new-mac-security-controls-as-ai-agents-raise-privacy-risks/ Bloomberg via Moneycontrol , “Apple is building its own AI agent and making it harder for rivals to control Macs,” November 20, 2026: https://www.moneycontrol.com/news/business/apple-is-building-its-own-ai-agent-and-making-it-harder-for-rivals-to-control-macs-14085248.html https://www.moneycontrol.com/news/business/apple-is-building-its-own-ai-agent-and-making-it-harder-for-rivals-to-control-macs-14085248.html Signal Just Called AI Agents an Existential Threat. Encryption Does Not Protect Your Screen. https://pub.towardsai.net/signal-just-called-ai-agents-an-existential-threat-encryption-does-not-protect-your-screen-8a3a64a70c5a was originally published in Towards AI https://pub.towardsai.net on Medium, where people are continuing the conversation by highlighting and responding to this story.