Hidden instructions for WhatsApp's AI help bot showed up in its help pages, including a rule against pointing users to privacy regulators. Meta blames a bug.
WhatsApp accidentally published the secret rulebook for its AI support chatbot. The text sat in the code behind about 80 help and FAQ pages on its website. One instruction told the bot not to recommend that users contact data protection regulators. Meta says a technical bug caused the leak and has removed the text.
A digital privacy firm called Redact analyzed the pages and found about 100 blocks of hidden text. The text was not visible on screen. It was tucked inside code tags labeled "HelpContentGalaxyLLMInclude". These hidden instructions are often called system prompts. They tell an AI model how to behave before it answers a single question.
The instructions covered disputes, account terminations and legal action. They included a section titled "WHAT NOT TO SAY". According to the report, the bot was told to:
WhatsApp spokesperson Joshua Breckman said the instructions were meant to keep support chatbots focused on technical product help, not legal interpretation. Meta says the text was backend guidance for the AI model and was never meant to be shown to users. Redact confirmed that all the affected code was removed from the public help center after it asked about the issue.
If you use an AI helper for account problems, expect it to stick to technical fixes. It is not built to explain your legal rights or point you to a complaint process. For those questions, a support chatbot is not the place to start.
The material does not say how long the text was public. It also does not say whether other Meta chatbots follow similar rules. The episode shows how companies are still learning to keep the rules for their AI tools separate from their public websites.
WhatsApp's AI help bot was told to stay on tech fixes, so do not count on it for guidance on legal rights or complaints.
A tech writer fed his Windows error logs to ChatGPT and learned his encrypted drive had no recovery key, a problem that could have locked him out for good.
Read more: ChatGPT read his laptop's error logs and found three hidden problemsA TikTok Shop affiliate says a strong first line, an honest demo and steady posting earned him six figures last year. Here is how it works and what it costs to try.
Read more: How a TikTok creator earns six figures with a 3-second opening lineGoogle's Gemini can now edit a PDF inside the chat, so fixing a travel plan no longer means downloading the file and starting over in another app.
Read more: Gemini can now edit your PDF itinerary without starting over
