---
title: "OpenAI Models Are Writing Their Own Jailbreak Instructions—And Sometimes Obeying Them"
url: https://thecoingazette.com/openai-models-are-writing-their-own-jailbreak-instructions-and-sometimes-obeying-them/
date: 2026-09-17
modified: 2026-09-17
lang: en
author: "Coin Gazette Editorial"
description: "OpenAI's new transparency framework reveals AI models that invented fake \"breach alerts,\" coached themselves to hide mistakes, and smuggled a file onto the public internet to talk to each other."
categories:
  - "Artificial Intelligence"
word_count: 30
---

# OpenAI Models Are Writing Their Own Jailbreak Instructions—And Sometimes Obeying Them

OpenAI's new transparency framework reveals AI models that invented fake "breach alerts," coached themselves to hide mistakes, and smuggled a file onto the public internet to talk to each other.