Meta Exposes Hidden Chatbot Instructions On WhatsApp Pages
Meta's WhatsApp inadvertently revealed internal instructions for its chatbot, which were not meant for public view. The exposure has since been fixed.
Original reporting · wired.com
Original: “WhatsApp Help Pages Leaked Hidden Instructions for a Meta Chatbot”
Read “WhatsApp Help Pages Leaked Hidden Instructions for a Meta Chatbot” at WIRED AI →
How this was written. Diagnosis Forward wrote this article with a language model from WIRED AI’s reporting (model: @cf/meta/llama-3.3-70b-instruct-fp8-fast); it describes that reporting rather than adding to it. The original, linked above, is the full story.
Key points from WIRED AI’s report
- Meta's WhatsApp exposed internal instructions for its chatbot due to a bug.
- The instructions provided guidance on how to respond to user inquiries and avoid providing legal advice.
- The exposure has been fixed, and the company has emphasized its commitment to providing transparent information about user data rights.
A recent analysis discovered that numerous WhatsApp help pages contained hidden text instructions intended for a large language model. These instructions were embedded in the webpage code and provided guidance on how to respond to user inquiries.
The exposed instructions included guidance on what not to say to users, such as avoiding legal advice or referencing specific data protection regulations. This information was not visible to users but could be accessed by viewing the webpage's underlying code.
According to WIRED AI, the instructions appeared in around 80 help and FAQ pages across WhatsApp's website. The text was removed after the issue was brought to Meta's attention, with the company attributing the exposure to a bug.
The hidden instructions provided insight into how chatbot responses are crafted and the efforts made to ensure accurate and relevant responses. The instructions also highlighted the distinction between the information provided to users and that given to large language models.
WhatsApp spokesperson Joshua Breckman stated that the instructions were intended to ensure support chatbots provide helpful responses to users. The company has since fixed the bug, and the instructions are no longer visible on the live pages.
The exposure of the hidden instructions raises questions about data protection and the handling of user inquiries. Meta has emphasized its commitment to providing transparent information about user data rights and support.
The incident has been resolved, but it may prompt further examination of how tech companies manage their chatbot responses and user interactions. The analysis by Redact, a privacy company, highlights the importance of monitoring and securing webpage code.
Questions and answers
What happened to the exposed instructions?
The instructions were removed after the issue was brought to Meta's attention, and the company attributed the exposure to a bug.
Why were the instructions hidden in the first place?
The instructions were intended for a large language model and not meant for public view, according to Meta.
How did the exposure occur?
The exposure occurred due to a bug that has since been fixed by Meta.
