New Delhi: Meta has removed hidden instructions intended for its artificial intelligence (AI) support chatbot after the text was discovered in the source code of dozens of WhatsApp help pages. The instructions included directions restricting the chatbot from recommending data protection authorities or discussing certain privacy laws when responding to users.
The findings, reported by privacy company Redact and subsequently covered by WIRED and India Today, revealed that around 80 WhatsApp help and FAQ pages contained hidden text designed for an AI system rather than human readers. Meta said the text appeared because of a software bug and that the issue had been fixed.
The incident has raised questions about how companies programme AI-powered customer support systems, particularly when users seek help with account deletion, privacy rights or possible legal action.
What did Redact discover on WhatsApp pages?
Redact examined 545 pages linked from WhatsApp’s help centre and identified 102 blocks of hidden instructions across approximately 80 pages.
The text appeared in the underlying website code under the heading “HelpContentGalaxyLLMInclude”. It was not normally visible to people reading the help articles through a web browser, but it could be found by examining the pages’ source code.
The instructions were intended to guide the company’s AI chatbot when answering support questions. While many blocks contained routine directions to help the chatbot provide relevant information, some included specific restrictions on how it should respond to questions involving legal matters and privacy regulations.
The discovery offered an unusual look at the instructions that can shape AI-generated customer support responses. It also highlighted the possibility of differences between information displayed publicly on a website and additional directions supplied to an AI system reading the same page.
Account deletion instructions raised concerns
One of the examples identified by Redact appeared on WhatsApp’s account-deletion help page.
The publicly visible information explained the timeframes associated with account removal after deactivation or inactivity. However, the hidden instructions added restrictions on what the chatbot should recommend to users seeking help with account deletion.
The text specifically told the AI not to recommend contacting a data protection authority, regulator or supervisory authority. It also instructed the chatbot not to suggest that WhatsApp could manually delete accounts on a user’s behalf. Such directions could be significant for users who have lost access to their accounts or are unable to complete the normal deletion process.
Data protection authorities are responsible for overseeing privacy laws in their respective jurisdictions. Preventing a support chatbot from recommending contact with these bodies could limit the options presented to a user, depending on the circumstances of the complaint.
The discovery does not establish that WhatsApp’s chatbot actually followed these instructions in conversations with users. Redact’s findings concerned the text exposed in the website code, and the extent to which it affected real support interactions remains unclear.
Hidden instructions restricted discussion of legal issues
Another set of instructions appeared on a WhatsApp page about reporting and blocking users.
The directions applied when a user mentioned legal action, lawsuits, lawyers, courts or litigation. They told the chatbot not to cite or interpret WhatsApp’s Terms of Service, Privacy Policy or other legal documents.
The instructions also restricted references to privacy and data protection regulations, including the European Union’s General Data Protection Regulation (GDPR), Brazil’s LGPD and California’s Consumer Privacy Act (CCPA).
Instead, the chatbot was directed to recommend consulting a qualified legal professional.
Meta’s position is that its support chatbot should help users with WhatsApp-related questions rather than provide legal advice. However, the discovery has prompted questions about where the line should be drawn between avoiding legal advice and providing basic information about users’ rights and available complaint channels.
The hidden text did not, by itself, prove that the company had violated any privacy law. It did, however, reveal how internal AI instructions can influence the type of information a chatbot is directed to provide.
Meta says a software bug exposed the instructions
After WIRED contacted Meta about the findings, WhatsApp removed the hidden instruction blocks from the affected pages.
WhatsApp spokesperson Joshua Breckman said the text had appeared because of a bug that had since been fixed. He explained that the instructions were intended for the company’s support chatbot and were not meant to be displayed directly to users.
The company said the guidance was designed to help the AI provide accurate and relevant answers to people who choose to contact WhatsApp Support through its AI-powered service.
Breckman also said WhatsApp provides transparent information about data rights and available actions. The company maintains that its support chatbots are intended to help users with product-related questions rather than provide formal legal advice.
The company did not publicly establish how long the instructions had remained in the pages’ source code before their removal.
Could other AI systems access the hidden text?
The discovery has also raised questions about how AI systems interact with publicly accessible websites.
Website source code can contain information that is not displayed in the normal page layout. Depending on how a website is built, automated systems and web crawlers may be able to read such information even when ordinary visitors cannot see it.
In this case, the hidden text was intended for Meta’s own support chatbot. However, the investigation did not establish whether third-party AI systems accessed the instructions or whether any external chatbot acted on them.
The incident illustrates a broader challenge for companies deploying AI-powered services: instructions meant for internal systems can become visible if website content is incorrectly configured.
It also shows why organisations need to check not only what users see on their websites but also what information is delivered through the underlying code.
What the incident means for WhatsApp users
The discovery does not mean that every WhatsApp support response was affected or that users’ private conversations were exposed. The reported issue involved hidden instructions on help-centre webpages, rather than evidence of a breach of encrypted personal messages.
Nevertheless, users seeking assistance with account deletion, privacy concerns or complaints may want to consult the relevant help pages and official regulatory resources directly instead of relying solely on an AI-generated response.
WhatsApp’s own help centre explains that AI-generated support messages may sometimes be inaccurate or inappropriate and advises users to report problematic responses and consult its help resources when necessary.
For issues involving legal rights or data protection, users can also seek guidance from a qualified professional or the relevant authority in their jurisdiction.
Greater scrutiny of AI customer support
AI chatbots are increasingly being used to handle customer questions, guide users through troubleshooting and explain how digital services work. Their answers, however, depend partly on the instructions, information and restrictions supplied to them.
The WhatsApp incident highlights the importance of ensuring that those instructions are consistent with publicly available information and that technical errors do not expose internal guidance.
It also underlines the need for clear boundaries between preventing a chatbot from offering professional legal advice and ensuring that users can still find accurate information about privacy rights and legitimate complaint procedures.
Meta has said the exposure resulted from a bug and that the hidden text has been removed. Whether the instructions affected actual chatbot responses remains unknown.
The episode serves as a reminder that AI support systems require careful oversight, transparent user information and regular technical checks as companies expand their use of automated customer service.