Google Messages Adds Optional Cloud Scam Checks and New Android Defenses
Google Messages will add an opt-in "enhanced protection" setting in the coming weeks. The setting adds a cloud-assisted check on top of the scam detection already on your phone, aimed at chats with senders you do not recognize. The Verge
When you enable it, Gemini models in the cloud study a stranger's intent across the conversation. Google describes this secure cloud review as a backup for on-device detection, meant to catch novel and complex attacks that smaller local models may miss. The goal is conversation analysis, not keyword matching. The system tracks how an unknown sender steers the dialogue over several turns.
Data sent for that review is encrypted and run in a secure, dedicated area that Google says is closed to Google and others besides the user. The design choice is practical. On-device checks keep data private by default and respond quickly. Cloud checks, or cloud inference, can run larger models with more context and update faster as new scam scripts appear. The cost is that data must travel, even when encrypted, which is why Google made it opt-in.
A second Messages control stays fully on the phone. Enhanced protection uses Gemini Nano, a compact model that runs locally, to examine live web clues in unfamiliar URLs before you open them and judge if the links are malicious. The message itself can look harmless while the link points to infrastructure set up hours ago. Checking live clues at tap time catches fast-flux domains, addresses that change rapidly to dodge blocks, and short-lived phishing kits better than static blocklists.
OS-level controls
The update goes beyond Messages into Android. Android is adding protection that stops apps from turning on call forwarding and alerts you when an app tries. Call forwarding abuse is a quiet persistence trick. Once switched on, your calls route to an attacker-run number, which can intercept return calls and one-time codes. Blocking that API path, the software route apps use to request the change, and showing the attempt removes quiet, programmatic control.
Android will also hide sensitive information automatically while you share your screen in Chrome and Android WebView, the component apps use to display web content. Remote-access and fake support scams often rely on you sharing your screen during login or payment. Automatic redaction during browser screen sharing lowers what that trick can collect, without asking you to spot the scam in the moment.
Google's chat-notification scam detection, introduced on Pixel phones last year, is expanding to Honor, Motorola, Oppo, OnePlus, Realme, Tecno, Vivo, Xiaomi and more devices. Google plans to bring that detection to hundreds of millions of additional users later this year. That expansion broadens coverage beyond first-party apps, because notification-layer checks catch scam text from other messaging apps at the point it appears to you.
Where this fits
The new controls build on layers Google has added over the past two years. Google Messages uses AI to flag text patterns linked to scams, including package delivery and job scams that make up much of current volume. Google Google has said its AI-powered Scam Detection blocks billions of suspicious messages for Google Messages. Chrome's Enhanced Protection uses Gemini Nano for added scam defense, while Chrome on Android has added AI-powered warnings against spammy notifications.
The threat is organized and AI-assisted. Scammers used Gemini AI to help draft spam messages, according to Google, with one group accused of sending 2.5 million messages to Android users over a two-week period. Bloomberg Alphabet's Google filed a New York lawsuit against makers of software allegedly enabling AI-powered phishing attacks. Reuters
Regulatory pressure is running in parallel. EU consumer groups filed complaints alleging Google, Meta and TikTok fail to remove fraudulent ads. India ordered removal of Google Firebase accounts after finding a scam pattern tied to Android malware posing as legitimate banking services. Singapore police ordered Apple and Google to prevent spoofing of government agencies on their messaging platforms.
The broader context here is a shift from filtering known-bad content to enforcing on intent and behavior. Blocking a listed URL or phone number scales poorly when new ones are cheap to generate. Studying conversational intent, checking URL clues at click time, denying call-forwarding APIs, and redacting screen shares all assume the lure may arrive. The focus moves to how much leverage an attacker keeps after delivery.
In my view, the opt-in design is worth flagging for enterprise and security teams. The users who would gain most from cloud-assisted intent checks are least likely to switch it on. Default-on with clear data handling would cover more at-risk users, but would draw closer review over message content leaving the device. Google has chosen explicit consent plus an isolated-processing claim to handle that trade-off. Uptake numbers, false positives on legitimate outreach from strangers, and the speed of the URL check will decide whether this becomes routine hardening or stays a power-user switch. Over the long arc, such trade-offs have tended to settle toward stronger defaults once the privacy design is proven, and this release sets up that path.
