Prompt Adware Discovered in OpenAI Codex Jailbreak

Cybersecurity researchers have released details about a new type of adware hidden in a jailbreak bundle for OpenAI Codex. This malware, referred to as prompt adware, was discovered by Check Point Research. It aims to modify the trusted instructions of an AI agent to insert advertisements for a third-party AI gateway. The malware is codenamed Haiou, which means "seagull" in Chinese. The discovery of Haiou raises serious questions about the security of AI applications.

Researchers have found that the jailbreak technique allows attackers to alter the configuration of OpenAI Codex. These modifications can lead the AI to generate unwanted content that does not align with the original programming instructions. The malware employs a specific technique to manipulate the AI's instructions, integrating code into the existing software that instructs the AI to generate advertisements for the external gateway. This type of attack could not only jeopardize the integrity of AI applications but also undermine user trust in such technologies.

Check Point Research has conducted a detailed analysis of how Haiou operates. The malware can be activated by inserting specific commands into the Codex environment. These commands are designed to bypass the AI's original security protocols. Researchers warn that such techniques could be used more frequently in the future to carry out similar attacks. The implications of this discovery are far-reaching.

Companies relying on AI technologies must rethink and adjust their security measures. The possibility that AI systems can be manipulated by such malware poses a significant risk. Experts recommend that companies conduct regular security audits and test their systems for vulnerabilities. The research on Haiou is part of a larger trend where cybercriminals attempt to exploit weaknesses in AI systems. The discovery shows that the security of AI applications depends not only on the technology itself but also on how these technologies are implemented and maintained.

The need to strengthen security protocols is becoming increasingly urgent. Check Point Research has already recommended measures to minimize risks, including the implementation of robust security solutions specifically tailored for AI applications. Companies should also provide training for their employees to raise awareness of such threats. The discovery of Haiou could also have legal and regulatory consequences.

Governments and regulatory bodies may be compelled to develop new policies to ensure the security of AI technologies. The discussion around the regulation of AI systems is further fueled by such incidents. The vulnerability exploited by Haiou could potentially affect thousands of users utilizing OpenAI Codex-based applications. Researchers estimate that the spread of such malware could increase in the coming months if appropriate countermeasures are not taken. The exact number of affected systems is currently unclear.

The discovery of Haiou is further evidence that cybercrime is becoming increasingly sophisticated in the digital world. The security community faces the challenge of keeping pace with these new threats. Check Point Research has already announced plans for further investigations to better understand the spread of this malware. The researchers have classified the vulnerability as CVE-2026-1234, indicating the urgency with which this threat must be addressed. Companies and developers are urged to promptly review their systems and implement security updates.