Android adds Gemini AI defenses against scam texts and calls
Google is rolling out new AI-based protections on Android that target scam messages and the abuse of call forwarding. The features are reaching eligible devices now.
The company says the changes respond to mobile scams that have become harder to recognize. Fraudsters now use spoofed phone numbers, fake bank alerts and AI voice cloning, not just basic robocalls and phishing texts.
"With scams growing harder to spot, it's nearly impossible for users to catch every red flag themselves," Google said.
According to Google, the scam detection already built into its Phone and Messages apps flags more than 5 million suspected scam calls and texts every month. The company expects the new measures to catch more.
Gemini checks conversations in the cloud
Most of the new protections are in Google Messages. For more complex conversations, the app can send messages to a cloud-based analysis backed by Gemini, Google's AI model, which looks for signs of fraud.
This analysis covers conversations with unknown senders. It runs inside Private AI Compute, an encrypted environment that Google says keeps user data private throughout the process.
There is also a local component. Gemini Nano, the smaller AI model that runs directly on the phone, checks unfamiliar links against real-time web signals. Users will also see clearer warnings before they open a link sent from a number that is not saved in their contacts.
The cloud-based protections are off by default, so users have to turn them on.
Changes across the wider system
Google has made several other changes to Android outside of Messages:
- On-device scam detection for chat notifications, which first launched on Pixel phones, will reach hundreds of millions more users through Android device partners later this year.
- Sensitive-information masking now extends to Chrome and Android WebView. Login fields and payment screens are hidden while the user is sharing their screen live.
- Apps can no longer quietly turn on call forwarding in the background. Android also shows a warning before manually entered MMI forwarding codes take effect. MMI codes are short dial strings, usually starting with symbols such as * or #, that change network settings like call forwarding.
Call forwarding matters because attackers can use it to redirect a victim's calls, including verification calls from banks or other services, to a number they control.
Verified bank calls reach more institutions
Google is also expanding its verified bank calls system, which it introduced earlier this year, to more financial institutions.
The system checks whether an incoming call really comes from a bank's call center or whether someone is spoofing the bank's number. If the check fails, Android ends the call automatically.
Google says that since June the feature has blocked more than 3 million spoofed calls for two Brazilian banks alone, Banco do Brasil and Itaú Unibanco. Revolut and other global financial institutions are also supported.
Google advises Android users to install the latest release, such as the October 2026 Android security update, and to keep their apps updated. When someone asks for sensitive information or wants an account-related action, users should confirm the caller's or sender's identity through a separate channel.
Why It Matters
These protections target social engineering, which is often where mobile fraud succeeds. Victims are frequently tricked into giving away codes, sharing their screen or forwarding calls, and no software flaw is needed. Blocking silent call forwarding and masking payment screens during screen sharing closes some of the steps scammers commonly rely on. Banking malware also remains active on Android, as the recent Hirudo campaign that swaps banking apps showed, so these features are only one layer of defense.
The move fits a wider trend of platforms adding AI-based scam and impersonation warnings to their products. Microsoft, for example, plans deepfake and impersonation alerts in Teams.
The opt-in design of the cloud analysis is a trade-off. It respects privacy concerns, but it also means many users may never turn the feature on. It is worth watching how many people enable it, and how quickly partner manufacturers ship the on-device features to phones other than Pixels.