NATHANIELPETTUSCYBER INTELLIGENCE
NEW POST DAILY
BACK TO CYBER NEWS, BLOG & ANALYSIS

Signal blocks Gemini and Siri AI screen access to protect private chats

Signal is enabling new screen-security controls to stop operating-system AI agents from reading or capturing private conversations. The protection is automatic on supported iPhones and activates when Signal detects AI screen access on supported Android devices, but Signal says important gaps remain.

By Nathaniel PettusCybersecurity, Linux/UNIX, OSINT, and privacy-focused analysis

What happened?

Confirmed facts: On November 20, 2026, Signal announced new protections intended to stop operating-system AI agents from capturing Signal conversations. Signal says Block Siri AI is enabled by default in Signal 8.37 or later on iPhones running iOS 27 or later. On Android, Signal 8.35.1 or later automatically enables Screen security when it detects an AI assistant trying to read the Signal screen; users can also enable it immediately in Signal's privacy settings. When active, the controls block screenshots and screen recordings inside Signal, prevent Signal content from appearing in the recent-apps view, and restrict on-device AI assistants from reading the screen. Signal previously enabled screenshot protection on Windows 11 to prevent Microsoft Recall from recording Signal messages. Reuters independently reported the announcement after working with Signal and the AI Now Institute on earlier tests showing that Meta's Muse agent could use broad operating-system permissions to retrieve Signal message content outside Meta's official connector path. Important limits and unknowns: Signal calls its remedy incomplete. It cannot reliably offer the same protection on macOS, it does not cover every Android device, and blocking screen capture does not prevent every possible access path, including notification content or data available through the device filesystem. Apple, Google and Microsoft did not immediately respond to Reuters about Signal's new criticism. The announcement does not establish that every AI assistant has read Signal chats, that Signal's encryption was broken, or that Signal's servers were breached.

How the technology works

Safe defensive explanation: end-to-end encryption protects a message while it travels between participants and while Signal's service relays it. Once the message is decrypted and displayed on an authorized device, the operating system must render it on the screen. An AI assistant with screen-reading, screenshot, accessibility, notification or filesystem permissions may be able to collect that visible content after encryption has already done its job. Signal's new controls use operating-system screen-security mechanisms to refuse screenshots and screen recordings and to obscure the app from recent-app previews. On supported phones, that can also prevent an AI assistant from receiving Signal's visible screen as context. This is an endpoint-permission problem, not a cryptographic break: the agent is attempting to observe content on a device where the user can already see it. The protection is strongest against screen capture and cannot by itself guarantee that another app, a compromised device, a notification service or the other participant's device will never expose the conversation.

Who is affected?

Who is affected: Signal users who enable or are offered operating-system AI assistants such as Gemini on Android, Siri AI on Apple devices or Recall-related features on Windows should review the new controls. The privacy impact extends to every participant in a conversation because one person's device can expose messages written by someone else. Users at elevated risk—including journalists, activists, government personnel, attorneys, healthcare workers, executives and survivors of abuse—should pay particular attention to assistant, accessibility, notification and screen-recording permissions. Mac users face a specific gap: Signal says macOS does not give it a reliable way to block screenshots used by agentic systems. The new settings do not mean Signal has become unsafe, that encryption no longer works, or that every supported AI feature sends chat content to a remote server. Actual handling depends on the device, assistant, permissions and account settings.

What should you do?

What should you do? Update Signal through the official app store or Signal's official desktop channel. On iPhone, keep Settings > Privacy > Block Siri AI enabled; Signal also offers a temporary 10-minute exception when you intentionally need a screenshot. On Android, open Settings > Privacy > App security > Screen security and enable it now instead of waiting for an access attempt. Expect screenshots and screen recordings inside Signal to stop working while protection is active. On macOS 27, Signal recommends adding Signal to the Excluded Apps section in System Settings > Siri; people with higher risk should consider disabling Siri AI and removing unnecessary accessibility, screen-recording and full-disk-access permissions from agent applications. On Android, review Gemini's screen-context and notification access, and disable access you do not need. On Windows, leave Signal's Screen security enabled and review Recall filters and snapshot settings. Do not paste sensitive messages into an AI assistant merely because the original conversation was encrypted. Organizations should inventory endpoint AI features, require explicit approval before agents receive screen, notification, accessibility or filesystem access, and give high-risk staff a separate communications policy. These steps reduce exposure but cannot protect against a compromised device or a conversation partner who intentionally copies content.

OPINION

My analysis

My Analysis: Signal is drawing the correct boundary. A private app should be able to tell the operating system and its AI agents that conversation content is off-limits. Users should not have to choose between useful device features and the privacy promises of the apps they depend on, especially when enabling one broad permission can expose information written by other people who never consented to AI processing. The deeper problem is that application-level encryption can be undermined after decryption by an operating system that treats every visible screen, notification and file as potential AI context. Signal's response is valuable, but forcing each app to improvise screenshot blocks is not a durable privacy model. Apple, Google and Microsoft should provide documented, enforceable controls that let developers prohibit agent access to screens, notifications and private app data without breaking accessibility or ordinary screenshots. Those controls should be off by default for sensitive content, auditable by the user and narrow enough to avoid expanding surveillance. Privacy should follow the data after it reaches the device, not end at the encryption boundary.

Why this matters

Cyber incidents often sound distant or overly technical. The important question is whether the same weakness, behavior, surveillance power, or exposure exists in systems you use. Facts and opinion are separated here so you can judge both clearly.