Google Messages Scam Protection Data: What Is Actually Sent

Google is rolling out an optional Enhanced protection mode for Google Messages that taps cloud-based Gemini models to evaluate complex conversation context from unknown senders. Unveiled in an October 14, 2026 Android Security blog post by Alberto Pastor Nieto and Shaifali Bhandari, the new feature tackles scams that develop gradually across multiple messages rather than relying on an obviously malicious opening text.

Private AI Compute Cloud Analysis Architecture

With the new setting enabled, Scam Detection can send message context to Google’s cloud Gemini models. Google says that This secure cloud analysis augments and extends our existing on-device detection, allowing us to identify novel and sophisticated attacks, actively uncovering and stopping abusers who try to hide their tracks. The company states this processing takes place inside Private AI Compute. According to Google’s documentation, remote attestation and encryption connect the device to a hardware-secured infrastructure designed so that processed data remains inaccessible even to Google.

However, important details regarding the pipeline remain undocumented. Google has not specified whether Enhanced protection transmits complete messages, selected excerpts, derived signals, or sender identifiers, nor has it published a retention period for the data handled through the feature.

Hardware Requirements Vary for AI Compute and Gemini Nano

The staggered rollout requires users to opt in, and compatibility varies depending on the hardware. A footnote attached to the Private AI Compute protection lists support for Google Pixel 6 and later, the Samsung Galaxy S24 series and later, Galaxy Z Flip6 and Z Fold6, and other Android AICore devices. Meanwhile, the localized Gemini Nano link checks demand newer hardware, specifically Pixel 10 and later, the Galaxy S26 series, and Galaxy Z Flip8 and Z Fold8 models.

These cloud features sit alongside other defenses. Enhanced protection also brings redesigned link warnings that slow users down with clearer context before opening links from numbers not saved in contacts. Google is expanding its chat notification scam alerts to hundreds of millions of additional devices later in 2026 through partnerships with manufacturers including Motorola, OnePlus, Xiaomi, and vivo.

Encryption Protects RCS Messages During Transit

End-to-end encryption protects RCS conversations while they travel between participants, but it does not dictate how messages are handled once they reach the recipient’s device. Enhanced protection introduces an optional analysis step that occurs after Messages has access to the received conversation.

Users must also distinguish between automatic background analysis and manual actions. As Google’s support documentation clarifies, reporting a conversation as spam sends the spammer’s number and recent incoming messages to Google for review and model improvement, which is a completely separate path from the automatic Private AI Compute evaluation.