Meta Rolls Out On-Device AI Tool to Detect WhatsApp Scams

Meta Platforms has introduced an optional security tool for WhatsApp designed to detect potential fraud by running machine learning algorithms directly on user devices, as first reported by The Verge.

The new capability, officially named Scam Alert, is currently being distributed through a limited beta testing program. When active, the system analyzes incoming conversations locally to determine if a message displays characteristics common to deceptive schemes.

If the local artificial intelligence model identifies a likely scam attempt, WhatsApp displays a caution notification within the conversation interface. This warning is completely hidden from the counterparty in the chat, allowing the recipient to privately evaluate the interaction without alerting the sender.

Upon seeing a notification, users are provided several immediate options, including blocking the sender, reporting the account, or proceeding with the communication. Should a user determine that a warning was generated in error, they can designate the conversation as a trusted chat.

Marking a conversation as trusted immediately clears the active warning and prevents Scam Alert from flagging future messages within that specific thread. Additionally, individuals who mark a thread as trusted are given the choice to voluntarily transmit the previous five received messages to WhatsApp engineers to refine the model's overall precision.

To address privacy concerns associated with automated message scanning, Meta emphasized that all data evaluation takes place locally on the handset. The company stated that no message contents are transmitted off the device for classification purposes, and no information is automatically forwarded to WhatsApp, Meta, or third parties. Users retain the ability to toggle Scam Alert off whenever they choose.

The introduction of Scam Alert follows Meta's earlier security update this year, which brought scam detection mechanics to WhatsApp device linking requests. The social media giant has been under pressure to fortify its global messaging infrastructure against increasingly sophisticated financial fraud.

With a global user base exceeding 3 billion people, WhatsApp has emerged as a primary target for bad actors orchestrating high-value frauds, such as unauthorized wire transfer solicitations and financial relationship schemes commonly known as pig butchering scams.

Financial losses attributed to these operations remain substantial. Data from the Federal Trade Commission indicates that consumers reported losing $425 million specifically to fraudulent schemes on WhatsApp in 2025, representing a significant portion of the broader $2.1 billion in total losses linked to social media scams that same year.