WhatsApp’s hidden code reveals AI chatbot’s forbidden topics and legal advice ban
WhatsApp unintentionally displayed internal instructions for its AI support bot on about 80 help pages, showing directives like “WHAT NOT TO SAY” and bans on giving legal advice. Meta said a bug caused the leak and the text has been removed.
Researchers at Redact.dev found roughly 100 hidden instruction blocks embedded in the source of around 80 WhatsApp help and FAQ pages. The directives, marked under “HelpContentGalaxyLLMInclude,” told the AI support bot to avoid recommending regulators, providing legal opinions, or citing WhatsApp’s terms, privacy policy, GDPR, LGPD, or CCPA. One example on the account-deletion page instructed the bot not to suggest contacting data-protection authorities.
After one outlet alerted Meta, the company removed the concealed text, with spokesperson Joshua Breckman blaming a bug and emphasizing that the guidance is for internal chatbot use only. The leak highlights how companies program AI to steer user interactions and raises questions about transparency. Meta did not disclose how long the instructions were exposed or whether third-party bots could have accessed them.
Why it matters
It shows how hidden AI instructions can be unintentionally exposed, revealing corporate limits on automated advice and privacy handling.
In this story
