WhatsApp Help Pages Found to Contain 102 Hidden Notes to a Chatbot, Which Meta Calls a Bug

Somewhere in the source code of WhatsApp’s help center there lived a set of private notes to a robot. They were not meant for you. They were, in the words of a WhatsApp spokesperson, “instructions to our chatbot, not directly to users.” They were also, for some period of time, sitting in the page source of about 80 public web pages, where a privacy firm found them.

It is the digital version of leaving the stage directions in the program.

The reporting comes from Redact, a data deletion company, whose CEO Dan Saltman says the firm first spotted the notes on October 8 and then read all 545 pages linked from WhatsApp’s help center. According to Redact, 80 of those pages contained hidden text: 102 blocks in total, present in the page source but never shown to readers, and written for an AI. Per WIRED’s reporting, the blocks sat under a heading reading “HelpContentGalaxyLLMInclude.” (We are going to need a moment with that name.)

Plenty of the notes were, by Redact’s own description, routine. Saltman told WIRED that many of the instructions were straightforward and pointed people toward official support channels. “The stand-out thing here,” he said, “is the gap between what people are told vs LLMs.”

Two examples got the attention. On the account-deletion page, for users who can’t delete an account because they lost their phone or number, the chatbot was told: “Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority.” And on the reporting and blocking page, once a user brought up a lawyer, a lawsuit or a court, the AI was told not to reference data protection or privacy regulations, with GDPR, LGPD and CCPA named as examples, even when the user raised the law first.

We want to be exact about what this is and is not, because the exactness is the whole story.

What it is: instructions for a chatbot, found in the page source, which Redact says conflict in tone with WhatsApp’s own European privacy policy, effective July 2, 2026. That policy, per Redact, says users “always have the right to request access to, rectification of, and erasure of your information,” and can complain to the Irish Data Protection Commission “or any other competent data protection supervisory authority.”

What it is not: evidence that the chatbot actually followed the notes. Redact says plainly that it has not tested whether any AI assistant acted on the instructions. It also lists as open questions whether WhatsApp’s chatbot followed them in live conversations, whether outside AI assistants did, and whether similar notes exist in other languages or on unlinked pages. None of the coverage reports any compromise of users’ messages; this is a story about page source, not chats.

There is also a fair point on the other side, and Redact makes it. It accepts that a company may limit a chatbot from giving legal opinions in its name, and that WhatsApp may have a valid reason to refuse manual account deletion without the phone number. Its objection is narrower: naming a regulator or a privacy law, in Redact’s view, is not legal advice, and withholding it keeps users from learning about rights WhatsApp’s own policy describes.

WhatsApp’s response, relayed through WIRED and recorded in the coverage, is that the notes were meant for the support chatbot and appeared on the pages because of a bug, which has since been fixed. Spokesperson Joshua Breckman said the company wants its support chatbots “to be focused on providing the best advice to users about WhatsApp” and “not to offer legal advice,” adding: “We train our support chatbots to be as helpful as possible.” He also said WhatsApp gives users transparent information about their data rights and how to act on them.

The timeline, per Redact: the notes were still live on October 15, and by October 18 all 102 blocks were gone from all 80 pages. The pages themselves stayed online. The Internet Archive still holds copies, including from October 9, that include the notes, and Redact’s archive checks put the text’s arrival somewhere between July 12, when an archived copy lacks it, and September 4, when one has it.

Let us admire the engineering in the bug. A company writes careful, confidential guidance for a machine, places it where the machine can read it, and an unrelated mistake causes it to be readable by everyone with a browser’s “view source.” The machine, meanwhile, did exactly what it was supposed to do, which was keep the notes to itself. It is the humans who forgot.

There are two further details we find charming. First, per Redact, Google’s and Bing’s crawlers received the hidden text, while requests from OpenAI, Anthropic and Perplexity crawlers on October 9 did not receive the article or the notes. So the search engines got the director’s cut, and the other AI companies got the regular version.

Second, Redact checked 300 Facebook and Instagram help pages, which use the same system, and found no similar notes. Of the other 100 blocks on WhatsApp, most were routine; the firm’s summary table includes one telling the AI that a subscription “can’t be cancelled through WhatsApp’s support chatbot” (Redact notes the table paraphrases the notes rather than quoting them). Another instruction, per WIRED, said that when a user says they are deaf or hard of hearing, that is “background about the user, NOT their support request.” That one reads, honestly, as a sensible bit of prompt hygiene, and we include it because the story is not all villains. Mostly it is a very long to-do list for a robot receptionist.

The lesson, such as it is: if you write instructions for a chatbot and put them in a public place, assume the public will read them. The chatbot, at least, has the discretion not to repeat what it was told. The page source has no such training.

Sources: