Hidden AI rules: WhatsApp help pages told chatbot not to point users to regulators
Around 80 WhatsApp help pages contained invisible instructions for an AI chatbot – such as not recommending a data protection authority to users. WhatsApp blames a bug that has since been fixed.
Symbolic image · Photo: Yuri Samoilov / Wikimedia Commons, CC BY 2.0 (cropped and resized)
About 80 WhatsApp help pages contained 102 blocks of instructions for an AI chatbot that were invisible to ordinary readers. They were discovered by Dan Saltman of privacy company Redact, who read and archived all 545 pages linked from WhatsApp’s help centre.
On the account deletion page, the bot was reportedly instructed not to recommend contacting a data protection authority or any other supervisory authority. Another block, meant to apply when users mentioned lawyers, courts or legal action, told it not to cite or interpret data protection regulations.
A WhatsApp spokesperson told Wired magazine these were instructions to the chatbot, not directly to users, and that their visibility was due to a bug that has since been fixed. By 18 October the blocks had been removed. Redact did not test whether the chatbot actually followed the instructions and checked only the English-language pages.