Dozens of hidden instructions meant for Meta’s chatbot, buried within help pages for its WhatsApp app, have been exposed. Among them, stern warnings against providing legal advice and guidelines on handling user mentions of data protection. These findings, uncovered by privacy company Redact, suggest a level of detail in chatbot training that goes beyond what users are told.

The leak raises questions about the extent of AI’s influence in our interactions online and the extent of control tech giants have over the information we receive. It’s a reminder that even in the age of AI, the lines between human and machine communication remain blurred.

According to Meta spokesperson Joshua Breckman, the text was mistakenly displayed due to a bug. He stressed that the instructions were meant for the chatbot, not for users, and were intended to ensure accurate responses. But the very existence of such detailed guidelines hints at the complex, often opaque, world of AI development and management.

Among the hidden text, there were explicit instructions to avoid mentioning specific upcoming features, even when users brought them up, as well as guidelines on dealing with sensitive topics like legal action or data protection. These instructions, while not intended for public view, provide a fascinating insight into the strategies tech companies use to steer AI interactions.