WORLDTECHMeta says a bug exposed WhatsApp chatbot's internal instructions
Meta says a bug caused around 80 pages of instructions written for its WhatsApp chatbot to become visible on a WhatsApp website for a period, according to a report by Wired. The material was meant for the chatbot only, not for the people who use it.
What became visible
For a while, anyone who visited a particular corner of WhatsApp's website could read parts of the guidance, Wired reported. Around 80 pages carried instructions that were meant for the chatbot's eyes only.
The most eye-catching passage is a heading that reads "WHAT NOT TO SAY." The exposed material was operating guidance for Meta's AI chatbot rather than the personal messages or account details of WhatsApp users.
Meta's explanation
Meta attributes the disclosure to a technical error and says a bug was to blame. The company has not linked the incident to any outside attack.
Reports on the incident come from Wired and Crypto Briefing, the latter of which carried the story under its own headline.
Why such guidance is sensitive
A chatbot does not settle its tone, subjects and limits from scratch. Companies supply it with guidance on what to say, which topics to avoid and where it must not go. A section titled "WHAT NOT TO SAY" effectively outlines a product's guardrails and what a company fears the product might do unchecked.
Knowing where those guardrails sit can make it easier to probe around them, which is one reason companies usually keep such material internal. WhatsApp, with its very large user base, is a particularly sensitive place for an AI assistant to operate.
Open questions
- How long were the instructions publicly visible on the website?
- Has Meta said whether it will change how such internal guidance is stored or published?