WhatsApp's help centre carried 102 blocks of text that never appeared on screen and were written for an AI to read, according to a report published today by Redact, a data deletion company, including one that told the AI not to recommend contacting any data protection authority to people unable to delete their accounts.

In Short

WhatsApp's help pages contained hidden instructions for a chatbot, and two of them told it not to mention privacy regulators or privacy laws such as the GDPR to people who were stuck or who brought up a lawyer. These rights are spelled out in WhatsApp's own privacy policy, so the hidden notes kept people from hearing about them in the one place they went for help. WhatsApp says the notes were meant only for its own support chatbot, showed up because of a bug, and have now been removed.

What Redact found

The findings come from a post by Dan Saltman, founder and chief executive of Redact, dated October 19, 2026. According to Redact, it first noticed the hidden text on October 8, 2026, and then read all 545 pages that WhatsApp's help centre links to, saving a copy of each. Eighty of those pages carried hidden AI text. Across them sat 102 separate blocks. By PPC Land's calculation, that means roughly 14.7 percent of the linked help pages contained at least one block.

Most of the blocks were mundane. Two were not.

The text was embedded in the help pages in a way that was never rendered on a reader's screen. According to Redact, the hidden content sat near an element named HelpContentGalaxyLLMInclude in the page source. The "LLM" in that label stands for large language model, the technology underneath chatbots. Anyone viewing a help page in a browser saw only the visible note; the extra paragraphs were sent with the page but never drawn.

Redact's post carries two dated updates. The first, dated October 18, says WhatsApp had removed the notes. Both notes quoted in the post were still present when Redact checked at 21:40 UTC on October 15. When it checked again at 02:08 UTC on October 18, "all 102 hidden blocks were gone from all 80 pages," according to the post. The help pages themselves stayed online; only the hidden text was taken out.

The second update, dated October 19, says WIRED had reported on the findings. According to that update, WhatsApp told WIRED that the notes were instructions for its support chatbot, the one people reach when they message WhatsApp Support with AI, and that they appeared in the pages because of a bug that has since been fixed. A WhatsApp spokesperson told WIRED the company wants its support chatbots "focused on providing the best advice to users about WhatsApp" and "not to offer legal advice," as quoted in Redact's post.

The account deletion note

WhatsApp only allows account deletion from inside the app, on the main phone, with the phone number still working. Losing the phone or the number blocks the self-service route. The help page on deleting an account, at faq.whatsapp.com/2138577903196467, carries a visible note covering that situation. It says deactivated accounts are deleted 30 days after deactivation and that inactive accounts are "generally deleted after 120 days (365 days for US)."

Underneath that visible note sat text a reader could not see. It repeated the waiting periods and then added three instructions. According to the wording Redact copied from the page on October 8, 2026, the hidden block read:

"Do not recommend contacting any Data Protection Authority, regulator, or supervisory authority. 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."

The effect, as Redact describes it, was that "a person who was locked out and wanted their account gone got one answer, which was to wait up to a year. The other routes were kept out of the conversation."

The second note sat near the end of the page on reporting and blocking, at faq.whatsapp.com/414631957536067. It was triggered, according to its own wording, "When a user mentions legal action, lawsuits, lawyers, attorneys, courts, or litigation." Seven instructions followed. The AI was told not to cite, quote or interpret WhatsApp's Terms of Service, Privacy Policy "or any legal document," not to make statements about WhatsApp's liability, not to provide legal opinions on the company's behalf, not to "engage in legal discussion or debate," and not to assert whether WhatsApp was responsible for any damages.

One line went further than the others. The hidden text instructed the AI: "Do not reference, cite, or interpret data protection or privacy regulations (for example GDPR, LGPD, CCPA), including when a user raises them directly." The block closed by saying that "users should consult a qualified legal professional."

"The last order held even when you were the one who brought the law up," the Redact post states.

What WhatsApp's privacy policy says

The core of Redact's argument is a comparison between the hidden notes and WhatsApp's own published privacy policy for the European region, effective July 2, 2026. According to Redact, that policy says users "always have the right to request access to, rectification of, and erasure of your information." The policy also says users can lodge a complaint with the Irish Data Protection Commission "or any other competent data protection supervisory authority," and it names the GDPR when it sets out that right.

Ireland's regulator is named because of the one-stop-shop mechanism, which routes cross-border cases to a single lead authority - for Meta and WhatsApp, the Irish DPC. WhatsApp's dealings with that regulator have reached the EU's top court before: PPC Land reported in March 2025 that an Advocate General sided with WhatsApp on whether decisions by the European Data Protection Board can be challenged directly, in a dispute that grew out of the DPC's transparency investigation.

Redact is careful on one point. The hidden deletion note did not tell the AI to stay silent about erasure requests as such. It told the AI not to suggest WhatsApp would delete an account manually, while the policy says users can always ask. The other two comparisons, Redact argues, "pull in opposite directions": the policy names a regulator and a law, the notes said not to mention either, the second one only once a lawyer or legal action had come up.

The rights themselves do not depend on WhatsApp. In the EU they come from Article 17 and Article 77 of the GDPR - erasure and the right to complain to a supervisory authority. "An instruction to a chatbot can't remove them," the post states. "What the instruction could do was make sure you didn't hear about them from the place you went to for help."

The case for WhatsApp

Redact's post includes a section setting out WhatsApp's side. A company has an interest in keeping its chatbot from giving legal opinions in its name, and pointing people towards a lawyer is a defensible instruction. "If the note had stopped at 'don't give legal advice', there'd be no story," according to Redact.

The post's objection is narrower. "Naming a law that protects you isn't legal advice, and neither is telling you a regulator exists," it states, noting that the privacy policy already does both. Redact also accepts that WhatsApp may have a legitimate reason to refuse manual deletion: without the phone number, the company may be unable to confirm who owns an account. That, according to Redact, "is a fair reason to say no to a request. It isn't a reason to keep the regulator out of the answer for someone whose request goes nowhere."

Redact's own position deserves a note. The company sells tools that let people bulk-delete posts, messages and data across services, so account deletion is its commercial territory. The underlying wording it quotes, however, is preserved in public archive copies that anyone can inspect.

Who was reading the notes?

WhatsApp's own support chatbot was the intended reader, according to Redact. One hidden block mentioned that chatbot by name, and WhatsApp has since told WIRED the notes were written for it.

Delivery was broader. According to Redact, WhatsApp's server sent the notes, together with the rest of each page, to the crawlers of Google and Bing. When Redact requested the pages on October 9 using the names that OpenAI's, Anthropic's and Perplexity's crawlers identify themselves with, the page came back without the article or the notes. Redact says it has not tested whether any AI assistant acted on the instructions.

That detail touches a question search engines have already addressed. Content visible to a crawler but not to a person resembles the pattern that search policies call cloaking. In February, Google's John Mueller and Microsoft Bing's Fabrice Canel warned that separate pages written for AI crawlers could breach cloaking rules, with Mueller asking why AI systems would want a page "that no user sees." Hidden text delivered to bingbot and Google's crawler in a help centre is a different case from a publisher building a separate markdown page, but the mechanism - one version for people, extra material for machines - is similar.

There is also a security dimension. Instructions planted in web content that an AI later reads are the raw material of prompt injection, in which text on a page steers an AI system's behaviour. Here the instructions came from the site owner, aimed at its own bot, rather than from an attacker. Whether any third-party LLM that ingested the pages through a search index took them as guidance is one of the questions Redact says it cannot yet answer.

Facebook's and Instagram's help centres run on the same system, according to Redact. It checked 300 of their pages and found no notes like these.

The other 100 blocks

Nearly all of the remaining blocks were housekeeping, according to Redact. They clarified terms people confuse, pointed questions to better pages, or described paid plans. Redact published summaries of twenty of them - its own paraphrases, not the exact wording - and sorted them into four kinds.

Eight were classed as holds back. After a report on the block and report page, the AI was told to say the matter is handled internally and to share no further steps, timeframes or process details. On account deactivation, for someone who has lost their number, the AI was told not to suggest signing up again, getting a new number or contacting the mobile provider; waiting, in the note's words as quoted by Redact, was "the ONLY correct answer." Other notes said contacting support cannot reverse a ban, that a ban with no review button in the app is final, that a WhatsApp Plus subscription cannot be cancelled or refunded through support, that an account cannot be recovered if the carrier cannot return the original number, and that requests to join recommended channels are not accepted.

Three were talking points. The most relevant to advertisers sat on the ad preferences page. According to Redact's summary, the AI was told to reassure people that personal chats are not changing and are not used for ads, and that a paid subscription can remove ads from Status and Channels in the European Region, where available, for people WhatsApp knows are 18 or over and who have added their WhatsApp account to Accounts Center. That matches the advertising structure Meta has described publicly. PPC Land reported in June 2025 that WhatsApp would show ads in Status using signals from connected Instagram and Facebook accounts, and in March 2026 that Meta's DMA compliance report said ads would arrive on WhatsApp Channels and Status in the EU. The other talking points covered a Meta Verified blank-screen bug and the statement that no feature lets anyone view another person's chats.

Seven fell into the category Redact labels "protects you." One told the AI never to suggest entering a false age or birth year, including when the person has said they are under 18. Another noted that someone scammed in a group probably wants to report it rather than delete it, since chat history can be evidence. Others covered two-step verification takeovers, which force the real owner to wait seven days; the futility of disabling two-step verification to fix login problems; the rule that a friend's number cannot receive a verification code; an unofficial app also called WhatsApp Plus; and an accessibility note stating that "I am deaf" describes the person, not an audio fault.

Two were housekeeping: one telling the AI not to recommend reinstalling the app to fix calling problems, and one noting that Arabic phrases for "verifying my account" usually refer to the paid blue badge rather than a phone number check.

Inconsistent waiting periods

Redact also found that WhatsApp's own pages disagree on how long an unused account survives. The delete account page says 120 days, or 365 in the US. The page on inactive account deletion gives a single figure with no US exception, and Redact says it has seen that page state 120 days in one country and 365 in another. For someone locked out of an account, the difference between four months and a year is the difference between the hidden note's advice working quickly and working slowly.

How long were the notes live?

The timeline is incomplete. According to Redact, Internet Archive copies of the delete account page from July 12, 2026 do not contain the note, while the copy from September 4 does. The archive's October 9 copy still shows the notes; its October 18 copy does not. Redact first saw the notes on October 8 and says it found no earlier reporting on them. Anyone can check the archived October 9 page by viewing its source and searching for the phrase "Do not recommend contacting."

Several questions remain open by Redact's own account: whether WhatsApp's chatbot followed the notes in conversations, whether outside AI assistants obeyed them, whether similar notes existed in other languages, and whether help pages not linked from the help centre carried more. "We read its instructions, not its answers," the post states.

Why this matters for marketers

Two strands of PPC Land's coverage meet here. The first is Meta's push to put AI between businesses and customers on WhatsApp. In June 2026, Meta launched Meta Business Agent, an AI tool that handles customer chats on WhatsApp, Messenger and Instagram, building on a version already used by more than one million businesses. Every such deployment runs on instructions. The WhatsApp case shows what those instructions can look like when written by a company with legal exposure, and how far they can drift from the company's published policies without anyone outside noticing. Brands running their own support agents on Meta's platforms face the same gap between what a bot is told and what a privacy policy promises.

The second strand is access to WhatsApp's AI layer itself. Meta barred general-purpose AI chatbots from the WhatsApp Business API from January 15, 2026, leaving Meta AI as the main assistant. The European Commission accused Meta in April of using fees to keep AI rivals off WhatsApp and, on June 9, ordered Meta to restore free access pending a final decision. Regulators are already examining who controls the AI conversations on WhatsApp; the content of those conversations is a separate question this report brings into view.

Privacy rights inside AI conversations are a growing concern for regulators too. The Belgian Data Protection Authority published a citizen guide in April setting out eight GDPR rights individuals can use against AI systems, and MLCommons this month catalogued 25 ways AI agents can mishandle personal data. An instruction telling a support bot not to name the GDPR "including when a user raises them directly" sits squarely in the territory those documents describe.

There is a transparency question as well. Leaked AI instructions have drawn scrutiny before - PPC Land covered a leaked Claude system prompt in 2025 that described how the assistant decides when to search. And Meta's documentation has come under close reading this month: PPC Land reported on October 18 that a clause letting Meta crawl business websites to train AI was still in the October 30 Commercial Terms despite a report that it had been removed. In WhatsApp's case, the instructions were exposed by what the company describes as a bug, and removed within ten days of Redact's first observation.

Is a chatbot that declines to name a regulator giving legal advice, or withholding it? WhatsApp's answer, through its spokesperson, is that it wants its bots "not to offer legal advice." Redact's answer is that naming a regulator was never advice to begin with. Neither the GDPR's Article 77 nor WhatsApp's own privacy policy has changed.

Timeline

Summary

Who: WhatsApp, owned by Meta, and Redact, a data deletion company whose founder and chief executive Dan Saltman published the findings. WIRED reported on them and received WhatsApp's response.

What: Redact found 102 blocks of hidden text on 80 of the 545 pages linked from WhatsApp's help centre, written for an AI to read. One told the AI not to recommend contacting any data protection authority to people unable to delete their accounts; another told it not to reference GDPR, LGPD or CCPA once a user mentioned legal action, even when the user raised those laws. WhatsApp says the notes were for its support chatbot and appeared because of a bug, and it has removed them.

When: Redact first saw the notes on October 8, 2026. Archive copies suggest they appeared between July 12 and September 4, 2026. They were gone by 02:08 UTC on October 18, and Redact published its report today, October 19, 2026.

Where: WhatsApp's public help centre at faq.whatsapp.com, with the hidden text served to Google's and Bing's crawlers. The rights at issue apply under the GDPR in the EU, where WhatsApp's privacy policy names the Irish Data Protection Commission.

Why: The notes kept public, accurate information about regulators and privacy law out of answers from WhatsApp's support chatbot, contradicting the company's own privacy policy. For businesses deploying AI agents on Meta's platforms, the case shows how bot instructions can diverge from published commitments, and how those instructions can surface to search crawlers.