MetaがWhatsAppの詐欺メッセージをAIで検出する「Scam Alert」機能を追加Meta adds AI screening to detect WhatsApp scams

Meta is rolling out an optional Scam Alert feature on WhatsApp that uses on-device machine learning to flag suspicious messages, a move aimed at curbing the fraud and social-engineering schemes that increasingly spread through messaging apps. Because WhatsApp counts more than two billion users worldwide, even incremental improvements to how the app surfaces potential scams could affect a large population of people who rely on it for personal and business communication.

The core idea is straightforward. When the feature identifies a message that carries the hallmarks of a scam, it appears to warn the recipient before they engage further, giving them a chance to pause, reconsider, or block the sender. Because the feature is described as optional, users are likely to have control over whether they turn it on, rather than having it enabled automatically for every conversation. The rollout is being handled in a limited fashion at first, which is a common approach for Meta when it introduces safety and machine-learning features so it can monitor performance and reduce false positives before a wider release.

A notable technical detail is Meta's emphasis on on-device machine learning. Running the detection model locally on a user's phone means the analysis happens on the handset rather than on Meta's servers. This design choice is significant for WhatsApp specifically, because the platform uses end-to-end encryption, meaning message contents are not readable by Meta in transit. On-device processing allows the app to scan for warning signs without the company itself reading the messages, an approach that aligns with WhatsApp's long-standing privacy positioning. It is a similar philosophy to how some competing platforms handle sensitive content classification, keeping the computation on the device to limit what data leaves the phone.

The Scam Alert feature builds on earlier work. Earlier this year, Meta launched scam detection for device-linking requests on WhatsApp, a safeguard designed to catch a specific fraud tactic in which attackers try to trick victims into linking their account to a device the attacker controls. Once linked, a bad actor can potentially read incoming messages or impersonate the victim to reach their contacts. By adding a broader message-level Scam Alert on top of that device-linking protection, Meta appears to be assembling a layered set of defenses rather than relying on a single checkpoint.

The timing reflects a wider industry pattern. Fraud that moves through encrypted messaging apps has become a persistent problem, ranging from investment and cryptocurrency schemes to job-offer scams and account-takeover attempts. Because such messages are encrypted, platforms cannot simply filter them at the server level the way email providers scan for spam, which makes on-device signals and behavioral cues more attractive as a countermeasure. Meta has previously introduced other protective prompts on its messaging products, including warnings when users are added to unfamiliar groups and safety notices in Messenger, and this feature extends that trend of nudging users toward caution at the moment they might be at risk.

For users, the practical value will depend on how accurate the detection proves to be and how the warnings are presented. Overly aggressive alerts risk desensitizing people or flagging legitimate messages, while overly cautious ones may miss genuine threats. Meta's decision to start with a limited rollout suggests the company intends to refine the model against real-world usage before expanding it. Broader availability, additional languages, and support across regions would be logical next steps, though the specifics of the wider timeline were not detailed in the initial announcement.

It is also worth situating this within Meta's broader push to embed artificial intelligence across its products. The company has been integrating machine learning into everything from content recommendation to its Meta AI assistant, and safety tooling is one of the areas where on-device models can deliver a tangible benefit without compromising encryption. As scammers increasingly use AI-generated text and more convincing impersonation tactics, defenses that operate directly on the device may become a more common feature across messaging platforms. For now, the Scam Alert feature represents an optional, incremental layer of protection that WhatsApp users can choose to adopt as it becomes available.