Google is preparing to put AI-based scam detection directly inside Android conversations, expanding the operating system’s role from carrying messages to judging whether a sender may be trying to deceive the recipient. Engadget reported on October 14 that a new opt-in setting in Google Messages will use Gemini models to evaluate a sender’s apparent intent and warn users when a conversation shows signs of a scam.
The example described in the report follows a familiar social-engineering pattern rather than an obviously malicious link. A stranger claims to have sent a photo of the recipient with a dog, is told that the number is wrong, and then tries to continue the exchange by asking for the recipient’s name. That sequence matters because the proposed protection is aimed at the evolving context of a conversation, not just a single suspicious word or attachment.
According to Engadget, Google says message analysis will take place through Private AI Compute, where the content is encrypted and processed in a protected environment that the company says is inaccessible even to Google. Separately, Gemini Nano will examine links on the device before a user opens them. The report does not independently test those privacy claims, so the architecture should be understood as Google’s description of how the system is intended to work.
The Gemini-powered setting will initially be limited to selected hardware, including the Pixel 10 and newer models, Samsung’s Galaxy S26 line, and the Galaxy Z Flip and Fold 8, along with other eligible devices. Google is also redesigning scam warnings so that users receive more context and additional time before opening links from numbers outside their contacts. Engadget says the opt-in setting is due to begin rolling out in the coming weeks.
Google is widening another layer of detection beyond Pixel phones. Scam warnings inside message notifications are expected to arrive on devices from Honor, Motorola, OPPO, OnePlus, realme, vivo and Xiaomi later this year, according to the report. Showing a warning in the notification means a user may be alerted without first opening the suspicious conversation, potentially moving the intervention earlier in the interaction.
The announcement also extends beyond AI classification. Later this month, devices running Android 15 or newer are set to receive automatic screen-sharing privacy that hides passwords, banking details and credit-card security codes when Chrome or an Android WebView is visible. The protection addresses a different part of the fraud chain: the moment when a victim has been persuaded to share a screen and sensitive information becomes exposed.
Another defense is scheduled for early December on Pixel phones and selected other devices. It is designed to prevent apps from secretly enabling call forwarding in the background and to warn users when someone tries to persuade them to enter an MMI forwarding code. Such a code can redirect calls, so blocking covert changes and flagging the manual step targets both software abuse and social manipulation.
Google has also deployed a bank-call verification program with financial institutions in Brazil. Android checks whether an incoming call is genuinely from a participating bank; if verification fails, the system ends the call and alerts the user. Google told Engadget that the feature has blocked more than three million fake calls since its June launch. The company has not said whether or when it will bring the tool to the United States, though it plans to work with additional banking partners.
Taken together, the measures show Android treating scams as a sequence of connected risks: an enticing message, a dangerous link, a request to expose a screen, or an attempt to redirect calls. Gemini is one component of that strategy, while device-level controls handle the moments when prediction alone would be insufficient. How well the system balances useful warnings, false alarms and privacy will become clearer only after the new protections reach users at scale.
Comments
Loading comments…