Meta's WhatsApp Scam Alert: AI-Powered Detection

What is Scam Alert?

Meta is launching an optional Scam Alert feature on WhatsApp that uses on-device machine learning to flag suspicious messages. Rolling out in a limited beta, this tool aims to help users identify potential scams directly within their chats.

When the on-device model identifies a message as a likely scam attempt, the user sees a warning in the chat, which is not visible to the other person. From there, the user can decide what to do: block, report, or continue the conversation. If they decide that a warning is incorrectly flagged, the user can mark the chat as trusted, in which case the warning is removed and Scam Alert will not flag that chat again.

If a user marks that they trust a chat, they can also opt in to share the last 5 messages received with WhatsApp to help improve the feature’s accuracy. This new feature builds on Meta’s earlier scam detection for device linking requests on WhatsApp.

How Scam Alert Works

Scam Alert uses an on-device machine learning model to analyze incoming messages on WhatsApp. When the AI model identifies a message as a likely scam attempt, the user sees a warning directly in the chat. This warning is not visible to the other person, ensuring the conversation appears normal to the sender.

From the warning, the user has clear choices: they can block the contact, report the chat, or continue the conversation. If the user believes the warning is incorrect, they can mark the chat as trusted. This action removes the warning and ensures Scam Alert will not flag that specific chat again in the future.

Additionally, if a user marks a chat as trusted, they have the option to share the last 5 messages received with WhatsApp. This data helps improve the feature’s accuracy over time, allowing the AI model to learn from real-world examples and reduce false positives.

User Controls and Feedback

When Scam Alert identifies a potential threat, users see a warning directly in the chat, which remains invisible to the other participant. From there, you have full control: you can block the sender, report the conversation, or choose to continue chatting. If you believe the warning is a mistake, you can mark the chat as trusted. This action immediately removes the warning and ensures Scam Alert will not flag that specific chat again in the future.

To further refine the system, users who mark a chat as trusted have the option to opt in to sharing the last 5 messages received with WhatsApp. This data is used solely to improve the feature’s accuracy, helping the on-device model learn from real-world conversations that were incorrectly flagged. This feedback loop is essential for reducing false positives over time, ensuring the tool becomes more precise without compromising your ongoing conversations.

Meta's Broader Anti-Scam Efforts

These new protections build on a series of recent moves by Meta to combat fraud across its platforms. Earlier this year, the company launched scam detection specifically for device linking requests on WhatsApp, targeting a common tactic where scammers trick users into linking their accounts to a device the attacker controls. This feature helps prevent unauthorized access before it happens.

Meta is not alone in this approach. Google Messages is using AI to detect scam texts, applying a similar on-device model to flag suspicious SMS content. Both companies are leveraging machine learning directly on the user’s device, which helps preserve privacy while still offering real-time protection. These parallel efforts highlight a broader industry trend toward proactive, AI-driven safeguards against social engineering attacks, marking a significant shift from reactive reporting to preventative screening in everyday messaging apps.