Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarThe Verge→ original

Signal blocks screenshots on Android and iOS 27 so Gemini and Siri can't read chats

Signal has disabled screenshots in its app on Android and iOS 27 so that the Gemini and Siri assistants can't read conversations from the screen. End-to-end encryption protects messages only in transit; on the display, the text is already open, and the AI sees it just as a person does. The block closes only one path: access through device files remains open. That's why app developers handling personal data should check what the system's AI layer can see.

AI-processed from The Verge; edited by Hamidun News
Signal blocks screenshots on Android and iOS 27 so Gemini and Siri can't read chats
Source: The Verge. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceThe Verge ↗
Editor-in-Chief
Key metric #1Users can turn screenshots back on in settings, but Signal will warn them about the risk
Key metric #2On macOS the protection can't be enabled because the system lacks the necessary tools
Key metric #3Last year Signal used the same approach to shield chats from Microsoft Recall

The Signal messenger has banned in-app screenshots on Android and iOS 27 so that AI assistants built into phones, Gemini on Android and Siri on iPhone, cannot read your conversations from the screen. The problem is that end-to-end encryption protects a message only on its way from sender to recipient. Once it appears on the display, it is ordinary decrypted text, and the operating system's assistant, which analyzes screen content, sees it just as you do. For an app whose core value is privacy, this is a serious hole, and Signal is closing it with the tools available to it.

What Signal actually did

The logic is simple. System AI assistants analyze what is shown on screen to better understand the user and carry out their requests. If an app prohibits screenshots, the assistant doesn't get an image of the conversation. That is what Signal switched on for Android and iOS 27.

Signal President Meredith Whittaker wrote on the company blog that agentic capabilities built into the OS "undermine Signal's ability to independently guarantee privacy." Users can turn screenshots back on in settings, but the app warns that this could expose the contents of the conversation, both yours and your contacts', to AI agents and undermine privacy guarantees.

Last year Signal used a similar tactic against Microsoft's controversial Recall feature, which periodically captures the screen and stores a history of activity. Back then, protected chats were also shielded from capture.

Why this is only partial protection

Whittaker herself calls the decision "incomplete." Banning screenshots covers one path but not all of them. An AI agent can reach app data in other ways, for example through the device's file system. On macOS, Signal currently cannot enable such protection at all: according to Whittaker, Google, Apple, Microsoft and others don't give developers simple ways to restrict AI access to data without seriously degrading the user experience.

Third-party agents are a separate issue. Signal points to the risk from assistants such as Muse from Meta, which could use deep accessibility settings as a "skeleton key." These features were created for people with visual or motor impairments and grant broad access to interface content. If such access is allowed, an agent can effectively read everything on an app's screen.

In addition, developers have little information about how agents actually obtain app data. Signal acknowledges that, for this reason, it cannot build a more reliable solution, and it asks Apple, Google and Microsoft to provide "simple technical tools and precise documentation."

What this means for developers and product managers

First, the usual threat model is out of date. Encryption in transit and at rest used to be enough. Now you have to account for the fact that the OS's AI layer reads the interface after decryption. This applies not only to messengers but also to banking clients, healthcare services and enterprise apps with confidential documents.

Second, platforms still lack a convenient setting along the lines of "don't show this app to the assistant." Teams have to resort to a blunt measure: banning screenshots altogether and giving up a familiar feature. This is a trade-off at the expense of convenience, and Signal is making it deliberately.

Third, check in advance which of your app's data is reachable through the accessibility APIs and the file system, and where possible mark screens with sensitive information as protected.

Whittaker calls the integration of AI agents into operating systems without such tools an "existential threat to app-level privacy" and describes Signal as a "canary in the coal mine": the problem concerns the entire ecosystem. The article contains no response from Apple, Google or Microsoft, so for now protection rests with individual teams. If platforms don't add explicit access settings for developers, the list of apps with screenshots disabled will most likely keep growing.

Story

Key Questions Answered

What Signal actually did?

The logic is simple. System AI assistants analyze what is shown on screen to better understand the user and carry out their requests. If an app prohibits screenshots, the assistant doesn't get an image of the conversation. That is what Signal switched on for Android and iOS 27.

Why this is only partial protection?

Whittaker herself calls the decision "incomplete." Banning screenshots covers one path but not all of them. An AI agent can reach app data in other ways, for example through the device's file system. On macOS, Signal currently cannot enable such protection at all: according to Whittaker, Google, Apple, Microsoft and others don't give developers simple ways to restrict AI access to data without seriously degrading the user experience. Third-party agents are a separate issue.

What this means for developers and product managers?

First, the usual threat model is out of date. Encryption in transit and at rest used to be enough. Now you have to account for the fact that the OS's AI layer reads the interface after decryption. This applies not only to messengers but also to banking clients, healthcare services and enterprise apps with confidential documents.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…