SourceHow We’re Building Scam Alert on WhatsApp With End-to-End Encryption and Verifiability Guarantees
This allows users to detect fraudulent messages from strangers while keeping their private conversations encrypted and invisible to Meta.
- The feature runs an on-device machine learning model to classify scam patterns without message content leaving the device.
- Model versions and manifests are signed by Cloudflare using Ed25519 keys and published on a public append-only transparency ledger to prevent targeted model delivery.
- Anonymous warning counts and user actions are processed via confidential federated analytics in Trusted Execution Environments (TEEs) with differential privacy.
- The feature is launching in limited Beta with an expanded Bug Bounty program for independent security researchers.