Meta confirms bug exposed hidden chatbot instructions on WhatsApp

Meta’s AI chatbot on WhatsApp has a rulebook. For a while, anyone visiting the right corner of WhatsApp’s website could read parts of it. Around 80 pages on the site exposed instructions that were meant for the chatbot’s eyes only, Wired reported. Meta says a bug was to blame. The most eye-catching detail is a heading that reads “WHAT NOT TO SAY.” What actually leaked The exposed material was guidance written for Meta’s AI chatbot. It was not intended for the people chatting with it. Around 80 pages carried the instructions, according to the report. Meta’s explanation is short: a bug caused it. The company attributed the disclosure to a technical error rather than to any outside attack. The material involved was the chatbot’s operating guidance, not the personal messages or account details of WhatsApp users. Why a chatbot’s rulebook is sensitive A chatbot doesn’t decide its own personality from scratch. Companies give it guidance on tone, topics to steer clear of, and lines it should not cross. A section titled “WHAT NOT TO SAY” is, by definition, a map of the guardrails. It reveals what a company is worried its product might do if left to its own devices. Knowing where the guardra...