WhatsApp's Hidden AI Instructions Exposed
Redact found 102 AI-directed instruction blocks across 80 WhatsApp help pages, including directives telling its support chatbot not to recommend privacy regulators or cite data-protection laws. WhatsApp removed the text after reporting and attributed its exposure to a bug. ([Redact](https://redact.dev/blog/whatsapp-help-pages-hidden-ai-instructions), [WIRED](https://www.wired.com/story/whatsapp-help-pages-leaked-hidden-instructions-for-a-meta-chatbot/))
This is less a prompt leak than a governance failure: reasonable guardrails against legal advice crossed into suppressing relevant rights information.
- –The hidden instructions were delivered in webpage code and exposed to search-engine crawlers.
- –They conflicted with WhatsApp’s European privacy policy, which acknowledges users’ right to contact data-protection authorities.
- –Avoiding legal opinions is defensible; refusing to mention regulators or privacy laws is materially different.
- –Developers should isolate model instructions, audit retrieved content, and test support bots against adversarial legal and privacy queries.
DISCOVERED
22m ago
2026-11-07
PUBLISHED
35m ago
2026-11-07
RELEVANCE
AUTHOR
Better Stack
