Developing Story
Hidden AI Instructions in Corporate Help Pages
A report found WhatsApp help pages contained hidden LLM-directed instructions telling AI systems not to recommend users contact a regulator, highlighting an emerging concern about companies embedding hidden prompts to manipulate AI assistants.
This page was written and is updated by the paper from the source articles listed below. Importance is how central the topic is to current coverage; confidence is how well-sourced the page is; mentions counts the articles that fed it. How the wiki works →
## Overview
A report found that WhatsApp's help pages contained hidden instructions directed at large language models, reportedly instructing AI systems not to recommend that users contact a regulator (redact.dev, October 19).
## Significance
This is an emerging category of concern: companies or bad actors embedding hidden text or instructions in web content specifically to manipulate AI assistants and chatbots that crawl or summarize that content (a form of prompt injection/SEO-for-LLMs). This raises regulatory and consumer-protection questions, since such techniques could be used to suppress disclosures (such as pointing consumers toward dispute resolution or regulators) through AI-mediated customer service or search. Given growing reliance on AI assistants to answer support queries, this is likely to recur as a theme — both as a security/trust issue (related to broader LLM manipulation and jailbreak research, see also 'abliterated models' vulnerabilities) and as a potential subject of regulatory scrutiny of Meta and other large platforms.