Meta has reportedly exposed internal instructions governing WhatsApp's AI-powered support chatbot, revealing how the company controls automated responses, particularly when users raise sensitive legal and privacy concerns.
The disclosure was discovered by privacy firm Redact and first reported by Wired. The instructions were embedded within WhatsApp's support-page source code, making them accessible to anyone inspecting the website.
Researchers identified nearly 100 concealed instruction blocks across approximately 80 help pages, marked with the identifier “HelpContentGalaxyLLMInclude.”
Some sections explicitly outlined “WHAT NOT TO SAY,” including directions preventing the chatbot from interpreting or referencing WhatsApp's legal documents when users mentioned potential legal action.
Although many instructions simply directed users toward official support channels, their exposure revealed how companies manage chatbot responses to minimise legal and reputational risks.
Meta reportedly removed the exposed instructions after being notified, highlighting the importance of protecting internal AI configurations and preventing unintended disclosure.
The larger concern extends beyond Meta. As enterprises deploy AI-powered assistants, hidden instructions can become security and governance vulnerabilities. Stronger prompt protection, secure implementation and independent audits are essential for maintaining digital trust.
