ChatGPT Mobile Voice Mode Reported Leaking Hidden System Instructions

On October 6–7, user VoidStateKate posted a series of reports about a rare anomaly in ChatGPT's voice mode: voice replies kept exposing the model's instructions/context that should have stayed hidden, and she repeatedly tagged OpenAI's official accounts to flag the issue.

Confirmed

Unconfirmed

Why it matters

System prompts are generally considered core configuration belonging to whoever deploys the model. If they can leak straight to users via voice output, internal product mechanisms could be exposed. The fact that the anomaly was limited to mobile and accompanied by the dual-process phenomenon suggests it may be a display or session-sync bug in the mobile client, which OpenAI should investigate.

2026-10-06 ~ 2026-10-07 · 5 related posts

Primary sources

2 near-duplicate retellings: VoidStateKate · VoidStateKate