Check Point finds Haiou jailbreak injecting ads into Codex

Check Point disclosed Haiou, a Chinese-language OpenAI Codex jailbreak that injects ads into agent responses via modified instructions.

Check Point researchers identified Haiou version 4.5.0, a Chinese-language jailbreak package for OpenAI Codex that injects promotional messages for the gateway zz.aiapi2025.top by modifying global agent instructions. The adware uses modular prompts and six security skills, includes an uninstaller, and is packaged with Ed25519 signatures and AES-256-GCM decryption. GBHackers and Cyber Security News say ads trigger on “hi” or “hello”; The Hacker News reports “Seagull” or “hello,” 77 GitHub forks, and a bundled Codex-X tool. The toolkit was found on a misconfigured server hosting Cobalt Strike, with logs of compromises of Chinese-speaking users’ computers. Researchers found no evidence the author engaged in cybercrime.