Today, Signal did something I did not expect to see from a messaging app in 2026: it started blocking screenshots inside its own app, on purpose, by default.
The reason is not the one you might guess. This is not about stopping people from saving a message they were sent. Signal's president, Meredith Whittaker, announced the change in a blog post today, November 20, and the target is clear: the AI agents that now live inside phone operating systems and read whatever is on your screen.
On iOS 27, Signal blocks screenshots inside the app and adds a setting called "Block Siri AI." On Android, it automatically switches on its Screen Security feature when it detects an AI assistant trying to read the screen. On Windows, it already did something similar last year to keep Microsoft's Recall from capturing its window. On macOS, Signal says it cannot reliably block screenshots at all, and instead tells users to add Signal to the excluded-apps list under System Settings, then Siri.
Read that list again. Four platforms, four different answers, and one of them is "we cannot do this." That is the real story here.
The encryption did not fail. The screen did.
I build an encrypted messenger. Mine uses ML-KEM-768 for key exchange and ML-DSA-65 signatures, the same post-quantum standards NIST finalized, and the whole point of that work is a simple promise: only you and the person you are talking to can read your messages.
Signal makes the same promise, and its encryption has not failed. Earlier this month, DeepMind's researchers formally verified the compiled code behind the Signal protocol, the PQXDH handshake and the Triple Ratchet, and found no new flaws in the parts they examined. The math holds. The protocol holds.
But here is the part every messenger builder learns sooner or later: encryption protects a message while it travels. It says nothing about the moment the message is on your screen, decrypted, in plain text, in front of your eyes. That moment is the endpoint, and the endpoint belongs to the operating system, not to the app.
For years, the main threat at the endpoint was someone physically looking over your shoulder, or malware taking a screenshot. Now there is a third one, and it ships with the phone: an AI assistant that scans your screen to "help" you, reading your messages as part of its normal job. Apple's Siri AI has an Onscreen Awareness feature. Google's Gemini Intelligence does the same kind of screen reading on Android. Both are built into the operating system at a level no app can fully control.
Signal's blog post puts it bluntly. These agentic AI integrations, Whittaker wrote, are "changing this environment in a way that's making it increasingly hostile to privacy. This is an existential threat to Signal, and our ability to responsibly uphold our mission."
Existential is a strong word. She chose it anyway, and I think she is right to. If the operating system hands your decrypted messages to an AI agent, then end-to-end encryption still works exactly as designed, and your messages still end up read by a third party. The guarantee becomes a technicality.
What Signal actually changed
The details matter, because they show how limited the tools available to an app developer really are.
On iOS 27, Signal blocks screenshots within the app by default and adds the "Block Siri AI" toggle. When Signal detects a screenshot attempt or an AI assistant trying to access the screen, users get a pop-up offering the option to disable the block. You can turn the protection off. Signal warns that doing so can expose your messages, and your contacts' messages, to these assistants.
On Android, the approach is reactive: Signal turns on Screen Security when it detects an AI assistant attempting to access the screen. Signal says it can only detect those attempts on some Android versions, so the protection does not work everywhere.
On macOS, there is no screenshot block. Signal says it lacks the tools to do it reliably. The advice is to exclude Signal from Siri yourself, in System Settings.
And Signal is upfront that none of this is a complete fix. Blocking screenshots stops one route into your chats. It does not stop an agent from reaching content already stored on the device through filesystem access. It does not stop third-party agents that arrive with broad permissions granted during setup, including accessibility permissions that let them interact with other apps. Signal's own reporting on this names Meta's Muse agent as an example of that problem: Reuters reported earlier this week that it was able to bypass its own official restrictions.
So this is a partial defense, and Signal says so itself. That honesty is worth noting. A company that wanted a marketing win would have announced "your chats are now safe from AI." Signal announced a screen block, listed its gaps, and asked the operating system makers for help.
What Signal is asking for
The second half of the announcement is a demand, aimed at Apple, Google, and Microsoft. Signal wants documented, reliable ways for developers to opt their apps out of AI data access and processing entirely: screen reading, notifications, filesystem access, and system APIs. It also wants more information about how these agents function, so developers can assess what data the agents are accessing and adjust their systems accordingly.
"We want to be very clear: it is a serious problem that Signal, and application developers beyond, lack robust tools to prevent AI agents from undermining application-level privacy," Whittaker wrote.
That is the part of this story that matters most to anyone building software. Right now, an app developer who promises privacy has no clean way to keep that promise once the operating system decides an AI agent is allowed to read the screen. The tools are inconsistent across platforms, partly undocumented, and on some platforms they do not exist at all. Signal, with all its reach and reputation, is working around the problem with screenshot blocks and settings toggles. A smaller developer has even less to work with.
Signal also makes a point about confidential computing, the idea that data can be processed in a protected enclave so it stays private even while being used. Its position is that confidential computing is not a complete answer here either. I agree with the instinct: an enclave protects data from the cloud provider, but it does not change the fact that the agent was given your messages in the first place. The question is not only where the data is processed. It is whether it should have been handed over at all.
What this means if you build messengers
I keep coming back to one uncomfortable fact: the hardest part of building a secure messenger was never the cryptography. The cryptography is the part with standards, test vectors, and reference implementations. The hard part is everything around it. Key storage. Identity verification. Metadata. And now, the screen itself.
When I added safety numbers to my own messenger, the fingerprint that lets two people check they are really talking to each other, I was solving an endpoint problem too. The protocol can be perfect and still fail if the person on the other end is not who you think they are. Screen reading is the same category of problem: the cryptography did its job, and the message still leaked, because the leak happened after decryption, in a place the protocol cannot reach.
Signal's move gives the rest of us a template. Block what you can block. Say plainly what you cannot. And push the platforms to give developers a real opt-out, because without one, every privacy promise an app makes is conditional on the operating system feeling generous that day.
What you should do
If you use Signal, update it to the latest version so the screenshot block is active. On iOS, check the "Block Siri AI" setting. On Android, look under Privacy for Screen Security. On a Mac, add Signal to the excluded-apps list under System Settings, then Siri. And think twice before turning the protection off for convenience: that toggle does not only affect your messages. It affects everyone who talks to you.
The bigger question is not about Signal. It is whether the companies that build operating systems will give app developers a real way to say no to AI agents reading their users' screens. Signal asked today, in public, with the strongest language it has used in years. The answer will decide what end-to-end encryption actually means for the next decade.
Sources: Signal's blog post, "Protecting your privacy from AI agents," November 20, 2026; Reuters, "Signal restricts screenshots to block AI agents from reading messages," November 20, 2026; Digital Trends, "Signal is fighting back against AI agents snooping on your private chats," November 20, 2026.
Top comments (0)