Android expands AI scam checks in Messages and calls

Android is rolling out a broader set of anti-scam features centered on Google Messages and call-related protections, with artificial intelligence playing a larger role. According to Google as reported by The Verge, a new opt-in “enhanced protection” mode in Google Messages will use Gemini to assess conversations with unrecognized senders and try to detect fraudulent intent earlier. Other announced measures include link checks, protections during screen sharing, and restrictions on call forwarding changes that scammers may abuse.

The overall pitch is clear: mobile fraud increasingly relies on long conversations, urgent prompts and social engineering rather than only obviously malicious links. The new approach is designed to combine older on-device scam detection with additional analysis for unfamiliar messages and suspicious URLs. Still, the details differ somewhat across reports, and some elements mentioned by secondary outlets are not described in The Verge’s account of Google’s announcement.

How Google says message analysis will work

The most significant change is the new enhanced protection mode in Google Messages. The Verge says it is opt-in and will roll out in the coming weeks. Once enabled, it uses cloud-based Gemini models to quickly evaluate the intentions of an unrecognized sender inside an ongoing conversation. Google’s rationale, quoted by The Verge, is that scammers often hide their motives in slow-burn exchanges that build trust before the fraud attempt becomes obvious.

Hipertextual describes the feature in similar terms, saying the system will monitor conversations with unknown senders and send chat context to Gemini so the AI can try to infer the other party’s real intention. Pplware likewise says the feature analyzes the full context of a conversation when messages arrive from unknown numbers. Across the three reports, the common point is that the tool is meant to look beyond single keywords and instead judge patterns in a broader exchange.

Google is also extending another line of defense already seen on some devices. The Verge reports that scam detection for chat notifications, introduced on Pixel phones last year, is expanding to devices from Honor, Motorola, Oppo, OnePlus, Realme, Tecno, Vivo, Xiaomi and others. Google says it plans to bring this to “hundreds of millions of additional users later this year.” Hipertextual and Pplware also say notification-based warnings will no longer be limited to Pixel devices, though they list slightly different brands.

Cloud AI, on-device AI and link warnings

The announced system appears to split work between cloud and on-device models. The Verge says cloud-based Gemini handles the higher-level assessment of an unrecognized sender’s intentions, while on-device Gemini Nano inspects real-time web signals in unfamiliar URLs before the user opens them, to help determine whether a link is malicious. Hipertextual describes the same structure: Messages would analyze links before they are tapped because scammers often create new websites that may not yet appear on traditional blocklists.

Two of the secondary reports add a further control around links from unknown numbers. Hipertextual says users will get a timed alert for such links, intended to slow down impulsive clicks and counter the sense of urgency often used in scams. Pplware similarly describes a deliberate delay with a timer for links from unknown contacts. However, this timed warning is not mentioned in The Verge’s report, so it should be treated cautiously unless confirmed directly by Google’s own announcement.

• Cloud-based Gemini is described as assessing the intent of unknown senders in conversations.
• On-device Gemini Nano is described as checking signals from unfamiliar URLs before they are opened.
• Chat notification scam alerts are expanding beyond Pixel to more Android brands.
• Screen-sharing sessions in Chrome and Android WebView will automatically hide sensitive fields.
• Android will block apps from enabling call forwarding and alert users if someone tries to do it.

Call protections and screen-sharing controls

Beyond messaging, Android is also adding safeguards tied to calls and remote-assistance scenarios. The Verge says Android will block apps from enabling call forwarding, while notifying the user when someone attempts to turn it on. This matters because call forwarding can be abused to intercept one-time codes or divert communications as part of an account takeover.

Hipertextual and Pplware add a more specific claim: if a user is persuaded to enter an MMI code manually to activate call forwarding, Android will warn the user before the change is applied. Both outlets also give timing details for this and other features, including references to late October and early December. Those dates do not appear in The Verge’s report, so they should be read as claims from those reports rather than fully corroborated facts across the source set.

Google also announced automatic concealment of sensitive information during screen sharing in Chrome and Android WebView. The Verge says sensitive information will be hidden while screen-sharing in those environments. Hipertextual and Pplware offer more detail, saying login fields and payment screens, including passwords or card security information, would be obscured from the person on the other side. They also say the measure applies to Android 15 and later, but that version detail is again absent from The Verge’s account.

Privacy promises and their limits

Because these features involve message content and web activity, privacy is central to Google’s presentation. The Verge says user data is encrypted during the AI analysis and processed in a “secure dedicated space” that Google, or anyone else besides the user, cannot access. Hipertextual echoes that claim, saying the analysis occurs in a private and secure space inaccessible even to Google employees. Pplware refers to processing through Private AI Compute and says the content is encrypted and handled in a strict environment.

Even so, the design still raises obvious boundaries and trade-offs. The enhanced message analysis is not universal by default: The Verge explicitly says it is opt-in. The cloud model is invoked for conversations with unrecognized senders, which means some contextual message data is being analyzed outside the handset, even if Google says the environment is isolated and inaccessible. The on-device link analysis appears narrower and may be easier to justify from a privacy perspective because, as described, it stays on the phone.

There are also limits in what can be concluded from the available reports. None of the sources provide quantified evidence of detection accuracy, false positives or false negatives. Nor do they explain in detail how users will challenge or override warnings, beyond the fact that the enhanced mode itself is optional. As with many anti-scam systems, stronger intervention may help against emerging fraud patterns, but users and organizations will still need to judge whether the privacy trade-off and the risk of occasional mistakes are acceptable.

Where the reports diverge

One source goes further than the others. Pplware says Google has also deepened partnerships with international banking entities to mitigate call spoofing and that Android could validate whether a call genuinely comes from a bank, rejecting the call if verification fails. That claim does not appear in The Verge or Hipertextual. On the basis of the provided sources alone, it should therefore be treated as an uncorroborated report rather than a confirmed part of the announcement.

For English-speaking businesses, the practical change is straightforward: customers and employees using Android may see more warnings around unknown-message conversations, suspicious links, screen sharing and call-forwarding requests. That could reduce some social-engineering attacks, but it may also add friction to legitimate outreach from unfamiliar numbers or support flows that rely on screensharing, so companies may need to make their communications clearer and prepare users for extra verification prompts.