GeneralBreaking NewsBreakingLatest News
Meta Accidentally Leaks WhatsApp AI Rules, Including What It Should ‘Not Say’ To Users
Sentiment: NeutralHigh Priority
What Happened?
WhatsApp unintentionally revealed internal instructions for its AI support chatbot, detailing how it should respond to sensitive inquiries about legal matters and privacy regulations. The exposed information was found in the underlying code of multiple help pages.
Key points
- WhatsApp exposed hidden instructions for its AI chatbot on dozens of help pages.
- Around 80 pages contained directives for a large language model (LLM).
- Nearly 100 blocks of hidden text were discovered, including 'WHAT NOT TO SAY' instructions.
- The chatbot was instructed not to reference legal documents or WhatsApp's Terms of Service.
- Meta removed the exposed instructions after being contacted about the leak.
Key insights
- The incident highlights potential gaps in AI training regarding sensitive topics.
- Users may be misled by AI responses that do not align with official policies.
- This exposure could lead to increased scrutiny of AI chatbot practices.
Why it matters
The leak raises concerns about how AI chatbots handle sensitive legal inquiries and user privacy.
Entities
Meta· OrganizationWhatsApp· ProductRedact· Company
Source
Times Now
