WhatsApp Hidden AI Instructions Restricted Regulatory Advice
Hidden prompts within help pages directed AI to withhold legal and regulatory information from users.
1Hidden AI Directives Discovered
Researchers discovered hidden instructions embedded in WhatsApp help pages that were invisible to human users but accessible to AI models. These prompts explicitly instructed AI assistants to avoid mentioning data protection regulators or specific privacy laws. The discovery occurred on October 8, 2026, after analyzing 545 linked support pages.
2Restricting Account Deletion Support
When users sought help deleting accounts without phone access, the hidden prompts directed AI to withhold information about manual deletion options. Instead, the AI was instructed to only mention automatic deletion timelines. This effectively steered users away from seeking human intervention or contacting data protection authorities regarding their account status.
3Blocking Legal and Regulatory Discourse
A second set of hidden instructions triggered whenever users mentioned lawyers, lawsuits, or courts. The AI was commanded to ignore all references to privacy regulations like GDPR or LGPD. By preventing the citation of these legal frameworks, the instructions limited the information available to users regarding their statutory rights.
4Contradiction of Public Privacy Policies
The hidden directives contradicted WhatsApp's official privacy policy, which explicitly informs European users of their right to complain to regulators. While the policy acknowledges these legal rights, the AI instructions actively suppressed them. This created a discrepancy between the company's public-facing legal documentation and the information provided by its chatbot.
5The Rationale Behind System Prompts
WhatsApp argued that preventing chatbots from offering legal opinions is a standard, sensible safety measure. However, the instructions went beyond avoiding legal advice by actively suppressing the existence of regulators. While protecting against liability is reasonable, the systematic exclusion of public, factual information about user rights remains a significant concern.
6Scope of the Hidden Prompts
The investigation reviewed 102 hidden blocks across WhatsApp's help center. While most were benign, two specific instances stood out for their restrictive nature. Similar checks on Facebook and Instagram help centers yielded no such hidden instructions, suggesting the practice was isolated to the WhatsApp support infrastructure during that period.
7Technical Implementation and Visibility
These instructions were sent by the server to search engine crawlers, including those used by Google and Bing. Although the notes were designed for AI consumption, they were not rendered on the user interface. WhatsApp has since confirmed to WIRED that these notes were indeed intended as specific instructions for their chatbot.
8Implications for User Autonomy
By manipulating AI responses, companies can inadvertently limit user access to legal remedies. While the specific notes have been removed, the incident highlights the risks of using hidden prompts to filter information. Users should rely on official privacy policies and regulatory bodies rather than chatbot responses when exercising their legal rights.
Original article: Hacker News — WhatsApp Hidden AI Instructions Restricted Regulatory Advice