Meta adds AI scam alert feature to WhatsApp in limited beta

Meta is rolling out an optional AI-powered feature on WhatsApp designed to catch scam messages before users fall for them. Called Scam Alert, the tool uses on-device machine learning to analyze incoming messages from non-contacts and flag anything that looks suspicious, all without shipping your chat data back to Meta’s servers.

The feature entered limited beta on August 12, 2026. It’s the latest move in what has become a sustained campaign by Meta to make its messaging platform less hospitable to fraudsters, following an earlier rollout of scam detection for device linking requests on WhatsApp.

How it works

When WhatsApp’s on-device model identifies a message as a likely scam attempt, it displays a warning directly in the chat. Only the recipient sees the alert. The person on the other end of the conversation has no idea it was flagged.

From there, the user has three options: block the sender, report them, or simply continue the conversation if they believe the warning was triggered in error. Users can also mark a chat as trusted, which prevents future false flags from the same contact.

The machine learning model powering the feature was trained on patterns drawn from previously reported scams. It analyzes communication patterns and linguistic signals to make its assessments. Crucially, the entire process runs locally on the device, meaning Meta never sees the content of your messages.

Meta says the implementation also employs advanced privacy techniques including differential privacy and Trusted Execution Environments, or TEEs. Differential privacy adds statistical noise to data so individual users can’t be identified. TEEs create isolated processing environments that prevent even the device’s operating system from accessing the data being analyzed.

Meta’s broader anti-scam push

In 2025, Meta removed over 159 million scam ads and millions of accounts linked to criminal operations. In the first half of 2026, WhatsApp took down almost 7 million accounts tied to various scam centers.

The earlier scam detection feature for device linking requests, implemented in March 2026, addressed a specific attack vector where bad actors would try to hijack WhatsApp accounts by tricking users into linking their accounts to unauthorized devices through deceptive QR codes or other fraudulent means. Scam Alert takes a broader approach, targeting the social engineering tactics that make messaging-based fraud so effective: fake investment opportunities, impersonation schemes, romance scams, and the like.

What this means for platform security

The feature is also notable for what it doesn’t do. It doesn’t automatically report flagged messages to Meta. It doesn’t block senders without user consent. It doesn’t override the user’s judgment.

The limited beta rollout suggests Meta is being cautious about false positive rates, allowing the company to refine the model before a broader launch. Meta hasn’t committed to a specific timeline for wider release.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.