Google Messages will use cloud AI to catch scams that build slowly over several texts

Google’s latest anti-scam move is best read as a shift from spotting suspicious messages one at a time to judging how an unfamiliar conversation develops. On 14 October 2026, the company announced Enhanced protection mode for Google Messages, an opt-in setting that can send message context to cloud-based Gemini models when users are talking with an unknown sender.

Google’s existing Scam Detection system in Phone by Google and Google Messages flags more than 5 million suspected scam calls and texts every month. The company is responding to mobile scams that use spoofed numbers, fake bank alerts and AI voice cloning.

In a survey Google conducted with YouGov across the US, Canada and Australia, 40% of smartphone users said they were “very” or “extremely” concerned about mobile fraud. Google also said iPhone users were significantly more likely than Google Pixel users to report falling victim to a scam and to describe their device as “not effective at all” against such threats.

The challenge for detection systems is the slow-burn scam. A fraudster may open with a wrong-number text, a supposed job offer or a friendly introduction, then seek money or account information only after several exchanges. Google says its larger cloud models can identify shifting intent in such conversations sooner than its existing local models.

Google had previously kept its scam-detection models entirely on Android devices to protect privacy. Enhanced protection changes that approach for users who turn it on, allowing Scam Detection in Google Messages to use Google’s cloud Gemini models to assess the context of a complex exchange with an unknown sender.

The company says the cloud analysis runs through Private AI Compute, where data is encrypted and handled in a secure dedicated space that is inaccessible to anyone else, including Google itself. The mode also combines on-device Gemini Nano with real-time web signals for link-safety checks.

The setting remains optional, and Messages can apply its cloud-based Gemini analysis only after a user enables Enhanced protection. Google has not published an independent accuracy rate for the feature.

Enhanced protection will roll out in the coming weeks on devices including Google Pixel 6 and newer, the Samsung Galaxy S24 series, Galaxy Z Flip and Fold 6, and other Android AICore-compatible devices.

Android is also rolling out other safeguards. All Android 15 and newer devices will receive automatic screen-sharing privacy protection this month, blanking login fields, passwords and payment screens. Android will also stop apps from silently enabling call forwarding in the background and warn users before a manually entered forwarding code takes effect.

Google is working with device manufacturers including Honor, Infinix, iQOO, itel, JOVI, Motorola, OPPO, OnePlus, realme, TECNO, vivo and Xiaomi to bring that call-forwarding protection to their hardware.