Hidden instructions in WhatsApp help pages told its AI chatbot not to point users to privacy regulators

WhatsApp help pages carried hidden directions telling the company’s AI support chatbot not to recommend data-protection regulators to people who were unable to delete their accounts, according to an investigation by Redact, a data-deletion software company. A separate instruction told the bot not to discuss privacy laws, including the GDPR, when users raised legal action. The text was removed from the pages by October 18, and there is no evidence in the report that the chatbot actually gave answers based on it.

The instructions were embedded in page code and were not visible to ordinary visitors. In Redact’s October 8 copy of the account-deletion page, the chatbot was told not to recommend contacting a regulator, asking WhatsApp to delete the account manually, or using another phone with an active number. The page’s visible guidance says account deletion must be initiated in the app on the account’s main phone, which leaves users who have lost access with limited options.

That direction sits uneasily alongside WhatsApp’s European privacy policy, which says users may complain to the Irish Data Protection Commission or another competent supervisory authority. The policy also says users can request erasure of their information. The hidden text did not tell the bot to deny those rights outright, but it did restrict the routes the chatbot could suggest to people seeking help.

The second block applied when a user mentioned lawyers, lawsuits, courts or litigation. It directed the chatbot not to cite or interpret privacy regulations, even if a user raised them, and to refer legal matters to a qualified professional. WhatsApp said the instructions were meant for people who choose to message its AI support service and that they appeared in help pages because of a bug. A spokesperson said the chatbot should provide relevant WhatsApp advice, not legal advice.

Redact said it found 102 hidden instruction blocks across 80 of the 545 pages linked from WhatsApp’s help center. Many were routine guidance for the chatbot, but the company also identified directions on account deactivation and reporting users that could narrow the answers people received. Redact said it found no comparable blocks in 300 Facebook and Instagram help pages it checked, even though those services use the same system.

The blocks were placed in HTML under the label HelpContentGalaxyLLMInclude. Redact found that Google and Bing crawlers received the text along with the rest of the pages, while its requests using the named crawler identities of OpenAI, Anthropic and Perplexity did not return it. Those checks do not establish whether any AI system ingested or acted on the instructions.

Redact found the notes on October 8 and published its findings on October 19. The text was still present on the live pages on October 15 and gone by October 18. The Hacker News report, citing WIRED, says WhatsApp removed it after WIRED contacted Meta. The company said the issue had been fixed, but the report says it remains unclear whether the same instructions still apply inside the chatbot.