WhatsApp's help pages hid LLM instructions: dont recommend a regulator
20 October 2026 · filed under efa8c4c666c6
Bulletin. WhatsApp’s online help pages contained hidden instructions directed at large language models, according to a report published by redact.dev and circulated via Hacker News on October 19. The instructions reportedly directed AI systems not to recommend that users contact a regulator.
The report does not specify which regulator or regulators were named in the text, nor does it detail how long the instructions had been present on the help pages or how many users’ queries may have been affected. Redact.dev’s account focuses on the discovery of the hidden text itself, embedded in material intended to be read by AI assistants processing support queries rather than by human visitors to the page.
The mechanism described follows a now-familiar pattern, text invisible or incidental to ordinary readers, placed specifically to steer automated systems that summarize or answer questions using that page as a source. In this instance, according to the report, the steering concerned a specific category of advice, whether to direct a user toward a regulatory body.
No response from WhatsApp or its parent company is included in the available material. The report does not state whether the instructions have since been removed. Readers are referred to the original redact.dev post for the full text of the instructions as published.