← AI PulseAug 12, 2026

Wire · news · Single-source brief

WhatsApp Introduces On-Device Scam Alert Feature

WhatsApp is rolling out an optional Scam Alert feature that uses an on-device machine learning model to identify potential scam messages while maintaining end-to-end encryption.

By Illumora Editorial

Source · Aug 12, 2026, 1:00 PM · On Illumora · Aug 12, 2026, 1:02 PM

Media from the primary source — shown here so you can stay on Illumora.

Rewritten from one allowlisted primary — not independent enterprise reporting. Lanes →

Brief drafted by Illumora’s editorial model from the linked primary source. Ops desk reviews flagged pieces. How we write →

Read the source →Meta Engineering (AI) — How We’re Building Scam Alert on WhatsApp With End-to-End Encryption and Verifiability Guarantees
Save

Meta Engineering announced an early look at Scam Alert, a new, optional feature for WhatsApp. This feature employs an on-device machine learning model to alert users to potential scam messages. The announcement was made on August 12, 2026, and the feature is currently in a limited Beta rollout.

WhatsApp states that this system is designed to protect user privacy, ensuring that no message content leaves the device for classification. The model operates entirely on mobile hardware, addressing previous challenges with performance, battery life, and model size for on-device classification.

Key Points

  • Scam Alert is an optional feature that users must enable.
  • A machine learning model is downloaded to the device and runs inferences locally.
  • The model classifies incoming messages from non-contacts based on known scam patterns.
  • Training for the model uses patterns observed in scam conversations from user reports.
  • If a likely scam is identified, a warning appears in the chat, visible only to the user.
  • Users can choose to block, report, or continue the conversation, or mark a chat as trusted.
  • Users can opt to share the last 5 messages with WhatsApp to improve accuracy if they mark a chat as trusted.
  • A confidential federated analytics pipeline, detailed in the USENIX NSDI 2025 paper "PAPAYA Federated Analytics Stack: Engineering Privacy, Scalability and Practicality," processes anonymous, differentially private aggregates.

Context

According to Meta Engineering, Scam Alert was designed with principles that adhere to the core guarantees of end-to-end encryption. The system's architecture reflects deliberate choices about its capabilities and limitations, with all inference occurring on-device. The feature is being rolled out in Beta, and Meta is working with its Bug Bounty community to stress-test the system, welcoming feedback from the broader security research community.

Why It Matters

This development indicates a shift towards enhancing user security against evolving scam tactics, including those using AI-generated lures, while prioritizing privacy through on-device processing and end-to-end encryption. It offers users a tool to identify potential threats without compromising the confidentiality of their communications.

What To Do

  • Note that Scam Alert is an optional feature and requires user activation.
  • Understand that message content remains on-device for classification.
  • Observe the options provided to users when a potential scam is flagged: block, report, or continue.
  • Consider the implications of marking a chat as trusted and the option to share messages for model improvement.

Keep Exploring

/techniques/multimodal-grounding