WhatsApp help pages hid text steering chatbot from regulators

WhatsApp's help pages contained hidden instructions aimed at the company's AI support chatbot, telling it not to recommend a data protection authority or regulator to users who could no longer delete their accounts, The Hacker News reported on 20 October 2026. The text was not visible to people reading the pages. It sat in the page code, and was removed by 18 October 2026.

According to the report, the hidden instruction sits awkwardly alongside WhatsApp's own privacy policy for Europe, which tells users they can complain to a data protection authority. A data protection authority is a national regulator that handles privacy complaints and can investigate companies. The help pages in question dealt with users who found they could no longer delete their accounts.

The case is unusual because of who wrote the text. Most reported cases of this kind involve an outside party planting instructions in a web page or document so that an AI system reads them and acts on them — a technique known as prompt injection, where text in content is treated by a model as a command rather than as information. Here, the instructions were on the vendor's own support pages, aimed at the vendor's own chatbot.

The mechanism is the same either way. Support chatbots are typically given access to a company's help articles so they can answer questions using current documentation. If those articles contain text addressed to the model rather than to the reader, the model may follow it. Readers of the page see nothing. The effect only shows up in what the chatbot says, or does not say, to the person asking.

The practical consequence described by The Hacker News is narrow but specific: a user who could not delete their WhatsApp account, and who asked the support chatbot for help, would not be told by that chatbot that a privacy regulator exists to take such complaints. The same user, reading WhatsApp's European privacy policy, would be told that they can complain to one.

The hidden text is no longer on the pages. The Hacker News reports it was removed by 18 October 2026. The report does not state who removed it, when it was first added, or how long it had been in place. It also does not say how many users interacted with the chatbot while the instruction was live, or whether the chatbot in fact followed it in every case.

For anyone running an AI assistant over their own documentation, the episode is a reminder that the content layer is part of the security and governance surface. Text that never appears on screen can still shape what an agent tells a customer. Any instruction placed in a knowledge base for a model to read should be treated as a published statement, because in effect that is what it becomes.