Haiou Adware Discovered Hiding in OpenAI Codex Jailbreak Toolkit

Cybersecurity researchers have disclosed important details regarding a newly discovered security threat targeting artificial intelligence tools. An OpenAI Codex jailbreak bundle was found to contain malicious code designed to manipulate trusted AI instructions.

Key highlights: - A prompt adware codenamed Haiou (Seagull in Chinese) has been uncovered by researchers. - The malicious bundle modifies the configuration and trusted instructions of OpenAI Codex. - Its primary function is to covertly insert advertisements for a third-party AI gateway into the AI agent's responses.

Understanding the Haiou Threat Experts from Check Point Research analyzed the phenomenon, classifying it as prompt adware. Instead of operating as intended, the compromised AI agent is manipulated through altered configurations to promote external commercial services. This technique demonstrates how attackers can exploit vulnerabilities in AI deployment pipelines and prompt structures to monetize traffic or redirect users without their explicit consent.

Implications for Developers As artificial intelligence integration grows, security awareness regarding prompt manipulation and untrusted third-party toolkits becomes crucial. Developers must carefully audit external bundles to prevent unauthorized modifications to AI workflows.