WhatsApp Adds Scam Alert: How the AI Guard Works

Meta has introduced a new security feature for WhatsApp called Scam Alert, designed to prevent fraud. A technical overview of the tool was published on August 12. The system’s primary goal is to warn users about suspicious messages before they fall victim to malicious actors.

How the Scam Alert Detection System Works

The entire verification process takes place directly on the user’s device. When the feature is activated, WhatsApp downloads a small AI model that analyzes incoming messages by comparing them against patterns of known fraudulent schemes. This includes phrases and communication structures typical of cybercriminals. Crucially, chat content is not transmitted to Meta’s servers, and end-to-end encryption remains fully intact.

Notifications and Control

If the algorithm detects suspicious activity, a corresponding warning appears in the chat. Notably, the sender of the message receives no notification that their actions have been analyzed. The user decides how to proceed: block the contact, report them, or simply continue the conversation. Meta only receives a notification if the user manually taps the “Report” button.

Feature Limitations: Are Contacts Safe?

The system has a significant limitation: the model only analyzes messages from users who are not in the device owner’s address book. This leaves a loophole for scammers using compromised accounts of friends or family. In such cases, Scam Alert remains inactive because the sender is considered a trusted contact.

AI Trust and Transparency

If the system flags a message as spam in error, the user can mark the chat as “trusted,” which stops further warnings from appearing. Meta has also prioritized transparency: each model version is recorded in a public registry with a corresponding checksum. Model weights are available to researchers to verify that it analyzes only fraud-related indicators, and system activity data is processed only in an anonymized and aggregated form.

When to Expect the Rollout

Currently, Scam Alert is in limited beta testing as part of the bug bounty program. Meta has not yet announced an official global release date or a list of countries where the feature will be available. Users are advised to follow standard security measures: if you receive requests for money or a phone number change, reach out to the person via a second, independent communication channel.