Check Point finds a Codex jailbreak kit that plants ads in the agent's own instructions

Check Point Research says a free jailbreak bundle for OpenAI Codex, called Haiou, rewrites the agent’s global instruction files so that greetings trigger an advertisement for a paid third-party AI gateway. The bundle also installs penetration-testing skills and hides its prompts in an encrypted package. Check Point calls the technique prompt adware and says it arrives as plain-language instructions rather than a program security teams would recognize. The repository has been forked 77 times and starred 403 times, according to The Hacker News. Check Point found no sign the advertised gateway is malicious and reports no confirmed victims. The article does not say whether OpenAI has responded.

Alan's takeNo surprise here. Free add-ons that promise to unlock your agent could be riddled with injections. History lesson: browser extensions were helpful, but often came loaded with unwanted functionality.