Prompt Adware Found in OpenAI Codex Jailbreak
Original source
Researchers Uncover Prompt Adware Hidden in OpenAI Codex Jailbreak Toolkit
The Hacker News →Security researchers at Check Point identified a jailbreak bundle for OpenAI Codex that persistently injects advertisements into AI model output. Named Haiou, it modifies configuration files to promote an opaque LLM gateway service. The toolkit employs prompt injection techniques—role-playing scenarios and CTF-style challenges—to bypass safety guardrails, while bundling penetration testing and reverse engineering capabilities. Common trigger words like “hello” activate the ad injection.
The “haiou-codex” GitHub repository has gained traction with 403 stars and 77 forks since launching in July 2026. It’s marketed as a localization tool for Codex and also packaged within Codex-X, a Chinese GUI configuration utility. Since Codex lacks official support in mainland China, the toolkit targets users seeking access to Western AI development tools.
The attack highlights a shift in how AI systems can be compromised: adversaries can inject undesirable behavior through configuration and natural-language instructions rather than deploying executable code. As enterprises adopt AI agents, the third-party skills and components governing their behavior become critical trust boundaries, creating fresh supply chain attack vectors that conventional detection methods may miss.
Read the full article
Continue reading at The Hacker News →This is an AI-generated summary. Read the original for the full story.