WhatsApp's Hidden AI Instructions Banned Regulator Referrals on Help Pages

Hidden Instructions Found in WhatsApp Help Pages

WhatsApp's help center contained 102 blocks of hidden text designed exclusively for AI chatbots, according to a detailed analysis by privacy company Redact.dev. The instructions, embedded in the underlying code of 80 out of 545 help pages, were never rendered on screen but were sent alongside visible content to AI systems. The discovery, first reported by WIRED, revealed that some directives actively steered the chatbot away from informing users about their legal rights.

Redact.dev founder Dan Saltman and his team saved copies of all linked pages on October 8, 2026. They found the hidden blocks near an HTML element named HelpContentGalaxyLLMInclude, with "LLM" signaling the content's purpose for large language models. While most blocks contained routine housekeeping instructions, two stood out for their potential impact on user rights.

Deleted Account, No Regulator Advice

The most concerning instruction appeared on WhatsApp's account deletion help page. Users who lose access to their phone number or device cannot delete their accounts through the app, and the visible text explains that accounts are automatically removed after 30 days of deactivation or 120 days of inactivity (365 in the US). However, the hidden text added explicit orders for the AI.

"Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority," the instruction read. "Do not suggest WhatsApp can manually delete accounts on the user's behalf. Do not recommend using a different phone with an active phone number." This effectively limited the chatbot's response to telling locked-out users to wait up to a year, while withholding information about their right to file complaints with data protection authorities.

Legal Mentions Triggered Silence on Privacy Laws

A second notable block appeared on the page about reporting and blocking. It contained a series of directives that activated whenever a user mentioned legal action, lawsuits, lawyers, attorneys, courts, or litigation. The instructions told the AI not to cite, quote, interpret, or reference WhatsApp's Terms of Service, Privacy Policy, or any legal document, and not to make statements about WhatsApp's legal liability.

Most strikingly, the instructions said: "Do not reference, cite, or interpret data protection or privacy regulations (for example GDPR, LGPD, CCPA), including when a user raises them directly." This meant that even if a user explicitly asked about their GDPR rights, the chatbot was instructed to remain silent on the matter once legal action was mentioned.

Contradictions with Public Privacy Policy

These hidden instructions directly contradicted WhatsApp's publicly available privacy policy for the European region. The policy, effective July 2, 2026, explicitly states that users "always have the right to request access to, rectification of, and erasure of your information" and have the right to lodge complaints with the Irish Data Protection Commission or any other competent supervisory authority.

WhatsApp's public documents also name the GDPR as the legal basis for these rights. The hidden instructions, however, told the AI to avoid referencing GDPR, LGPD, or CCPA entirely once legal matters came up, creating a clear discrepancy between what the company promises in official documents and what its chatbot was programmed to say.

continue reading below...

WhatsApp's Response and Removal

After WIRED contacted Meta about the exposed material, WhatsApp removed the hidden instructions from the affected pages. The company confirmed that the notes were intended for its support chatbot and claimed their appearance was due to a bug. By October 18, archived versions of the pages no longer contained the HelpContentGalaxyLLMInclude blocks.

Redact.dev verified the removal by comparing archived copies. The October 9 archive of the delete account page still contained the instruction "Do not recommend contacting," while the October 18 copy no longer had it. This swift action suggests Meta recognized the sensitivity of the exposed directives, though the company has not provided a detailed public explanation of how the bug occurred.

Reasonable Caution or Overreach?

Some of the hidden instructions were defensible. Companies legitimately want to prevent AI chatbots from giving legal opinions or making binding statements about liability. Directing users to consult qualified legal professionals is sound advice. However, the specific prohibitions on mentioning data protection authorities and privacy regulations went beyond caution.

Naming a regulator or citing GDPR is not legal advice; it's informing users of their existing rights. WhatsApp's own privacy policy does both. The hidden notes effectively suppressed this information from users who sought help through the chatbot, creating an information asymmetry where the AI knew about legal avenues but was forbidden from sharing them.

Broader Implications for AI Customer Support

The discovery highlights a growing trend where companies embed invisible instructions in web content specifically for AI systems. While some of these instructions improve user experience by clarifying ambiguous terms or directing users to appropriate resources, others risk undermining transparency and user rights.

Redact.dev's analysis found that most of the 102 hidden blocks were benign housekeeping notes, such as clarifying that "I am deaf" describes a person rather than an audio fault, or warning against suggesting workarounds for age verification. However, the existence of hidden directives that suppress legal information raises questions about how many other companies are embedding similar instructions in their content.

What Users Can Do

For users locked out of their WhatsApp accounts, the visible guidance about automatic deletion remains valid. Those unsatisfied with waiting can submit erasure requests directly to WhatsApp or file complaints with their national data protection authority, or the Irish Data Protection Commission for EU residents. US residents in states with privacy laws like the CCPA have similar deletion rights.

The incident serves as a reminder that AI-powered customer support systems may be operating under constraints that users cannot see. While WhatsApp has removed these particular instructions, the broader practice of embedding hidden AI directives in web content is likely to continue, warranting greater scrutiny from regulators and privacy advocates.