Free Codex Jailbreak Hides 'Prompt Adware' That Pushes Ads Through Your AI Assistant

Check Point Research has disclosed details of a jailbreak bundle for OpenAI Codex called Haiou (Chinese for Seagull). It changes Codex's configuration files and the agent's global instructions so that later responses include advertisements for the author's website. Researchers have named this behaviour "prompt adware". Common greetings and activation words, such as "hi", "hello" and "Seagull", trigger an ad for a website that promotes an opaque, OpenAI-compatible LLM gateway. That is a middleware service sitting between an application and an AI provider.

The bundle tries to defeat the model's safety training with prompt injection techniques, including role-playing and reframing cybersecurity requests as Capture The Flag scenarios. It also installs skills focused on penetration testing, cracking and reverse engineering. Check Point believes the likely business model is to use the freely shared jailbreak as a way to attract customers to the paid gateway.

On GitHub the package is described as a "local enhancement configuration package". Since appearing on July 10, 2026, the repository has been forked 77 times and starred 403 times. Codex is not officially available in mainland China, which suggests the tool targets people wanting to try Western AI coding assistants. The case shows that the instructions guiding an AI agent can be altered persistently, and that unofficial add-ons can change its behaviour in ways users do not expect.