WhatsApp Help Pages Told Its Chatbot Not to Name Data Regulators

Privacy company Redact found 102 hidden blocks of chatbot instructions in the source of 80 WhatsApp help pages, among them an order not to recommend contacting "any Data Protection Authority, regulator, or supervisory authority." Anyone could read them with a browser's view-source command, no account needed. They were gone by 02:08 UTC on Oct. 18.

Redact's chief executive, Dan Saltman, wrote in a blog post that he found the text on Oct. 8 and then saved all 545 pages linked from WhatsApp's help centre. The blocks sat in an element named HelpContentGalaxyLLMInclude. Most were routine; two were not.

The two blocks that mattered

The first is on the account-deletion page. It addresses a user who cannot delete an account because the phone or number is lost, and tells the bot not to recommend a regulator, not to suggest WhatsApp can delete accounts manually and not to recommend using a different phone with an active number.

The second is on the reporting and blocking page. When a user mentions lawsuits, lawyers or courts, it says "Do not reference, cite, or interpret data protection or privacy regulations", including when the user raises them directly. It closes by sending users to "a qualified legal professional."

WhatsApp's European privacy policy, effective July 2, 2026, says users "always have the right to request access to, rectification of, and erasure of your information" and may complain to the Irish Data Protection Commission, "or any other competent data protection supervisory authority." Saltman argues the hidden text conflicts with that, while accepting that a company can reasonably keep a chatbot from giving legal opinions.

What WhatsApp said

InfoSec Today republished WIRED's report, which quotes WhatsApp spokesperson Joshua Breckman as saying the text was displayed because of a bug that has been fixed. "These are instructions to our chatbot, not directly to users," Breckman said, adding that they are meant to ensure accurate and relevant responses for people who message WhatsApp Support with AI. The Terminal has not seen WIRED's original page.

Date	Event
July 2, 2026	EU privacy policy takes effect
July 12, 2026	Internet Archive copy of the deletion page has no note
Sept. 4, 2026	Archive copy includes the note
Oct. 8, 2026	Redact finds the text and saves 545 pages
Oct. 15, 2026, 21:40 UTC	Notes still present
Oct. 18, 2026, 02:08 UTC	All 102 blocks removed
Oct. 19, 2026	WIRED publishes

Source: Redact, Oct. 2026.

The archive dates bound the exposure to between July 12 and Oct. 18, roughly 14 weeks, though neither Redact nor InfoSec Today says how long the text was live.

What nobody has shown

Redact read the instructions, not the chatbot's answers, so it cannot say whether the bot followed them. It also cannot say whether other AI assistants that read these pages acted on the text. It requested a page using the OpenAI, Anthropic and Perplexity crawler names, and got the page back without the notes; Google's and Bing's crawlers got the full content.

Facebook and Instagram help centres use the same system. Redact checked 300 of their pages and found no similar notes. Of the other 100 WhatsApp blocks, some hold back detail on account bans and subscription cancellations, and some protect users, such as advice to keep chat history when reporting scams.

For another case of a privacy claim tested against what a system does, see the Terminal's story on Citizen Lab's Candiru ad-targeting findings. The week's other disclosures are in the Terminal's count of October Patch Tuesday flaws. The Terminal has found no report of a regulator asking WhatsApp about the text, and Redact lists other languages and unlinked pages as unchecked.