Check Point Research, the threat intelligence arm of the Israeli security company, published an analysis on November 3 of "Haiou," a Chinese-language jailbreak bundle for OpenAI's Codex coding agent. A jailbreak bundle tries to remove a model's safety limits. Check Point found the package on a misconfigured server run by a threat actor, next to hacking tools such as Cobalt Strike. It says there is no evidence linking the bundle's author to cybercrime. The bundle changes the agent's configuration so that replies carry promotional text for a third-party AI gateway. Check Point calls this "prompt adware."
The risk for companies is control. Developers who install such bundles let a stranger rewrite how their coding agent behaves and route attention to an unvetted service. Check Point says it saw no sign the gateway is malicious, but notes research linking such gateways to data-snooping risks. The package had more than 250 GitHub stars. Its Claude Code version contained no ad injection, Check Point says.
