WhatsApp help pages exposed hidden instructions for Meta chatbot

In short: Around 80 WhatsApp help pages accidentally revealed behind-the-scenes instructions meant for Meta’s support chatbot, and Meta says a bug caused it.

What happened

A new analysis found that about 80 help and FAQ pages on WhatsApp’s website included hidden text that looked like instructions for a chatbot. These instructions were not shown in the normal page text, but they appeared in the page’s code, which is like the “behind the scenes” layer a browser uses to display a webpage.

The findings came from Redact, a privacy company that scans big tech websites. Redact says it found around 100 blocks of hidden text across the pages, including sections labeled “WHAT NOT TO SAY.” Some instructions told the chatbot to avoid giving legal advice, and to avoid citing legal documents like WhatsApp’s Terms of Service.

On one page about deleting an account, the hidden instructions included a line saying not to recommend contacting a data protection authority or regulator. Another page about reporting and blocking included guidance for what to do if a user mentions lawsuits or lawyers. Other snippets included notes like treating “deaf or hard of hearing” as background, not the actual support request.

After WIRED contacted Meta, the hidden text was removed. A WhatsApp spokesperson said it was visible due to a bug that has now been fixed, and that the text was intended only for its support chatbot.

Why it matters

Many people assume help pages show the full story. This incident shows there can be extra rules running in the background, especially when companies use AI chatbots for support. Even if the text was not private user data, it can affect trust by showing a gap between what people see and what a chatbot is told to say.

Source: Wired