Tool requires users to opt in and runs entirely on the device
- Bookmark
- CommentsGo to comments
The new tool uses artificial intelligence to check through messages from people who aren’t in a user’s contacts. It looks at “conversational structure and linguistic signals” to try and understand whether a message is likely to be part of a scam.
If WhatsApp thinks that it is, it will show a warning in the chat that will not be visible to the other person. The user can then decide whether to block or report the user sending it – or carry on as normal. The tool has to be manually switched on. When it is, the app downloads a machine learning model that can check those messages – and is run entirely on the device, meaning that none of the conversation is sent to WhatsApp or to another third party, parent company Meta stressed.
Some information is shared with WhatsApp. To ensure that the feature is working – and to ensure that it stays up to date with new scams – the app gathers particular signals that are confidently processed and are made available entirely anonymously to Meta, it said.
The Scam Alert system is built using previous conversations that WhatsApp users have reported as possible scams. Those conversations are then used to build an AI-powered system that can identify whether a new message is similar, and therefore likely to be illegitimate.
For now, the feature is rolling out in a limited beta version, to early users. The company will “stress-test” the system and has encouraged researchers to try and find bugs within it, before it will be rolled out to all WhatsApp users.
Join our commenting forum #
Join thought-provoking conversations, follow other Independent readers and see their replies
Comments