Researchers Uncover Prompt Adware Hidden in OpenAI Codex Jailbreak Toolkit

Check Point Research identified a malicious OpenAI Codex jailbreak toolkit named Haiou, which modifies AI agent configurations to inject advertisements into responses. The jailbreak uses prompt injection techniques to bypass safety guardrails and persistently promotes a third-party LLM gateway at zz.aiapi2025[.]top. Distributed via GitHub as a 'local enhancement package' and bundled with Codex-X, the toolkit introduces supply chain risks by altering trusted instructions through natural language rather than traditional malware payloads.

AI-extracted · verify before operational use