Khaberni - The study carried out by the security firm "CrowdStrike" has proven that the "DeepSik" artificial intelligence model generates codes full of security vulnerabilities when linked to politically sensitive topics for the Chinese government, according to a report by the technical site "The Hacker News".
This study comes following a series of critiques directed at the model due to security concerns, as investigations found that the model avoids discussing topics sensitive from the perspective of the Chinese government, leading to its ban in several countries around the world.
The Taiwanese National Security Office has also warned its citizens against using Chinese AI models, including "DeepSik", fearing the distortion of historical narratives and the amplification of misleading information related to the tense relations between China and Taiwan.
The site reported that "the five Chinese artificial intelligence models have the capability to create scripts for network attacks and exploit codes for security vulnerabilities that enable the execution of remote code under certain conditions, increasing the risks of cybersecurity management."
CrowdStrike's study clarified that "DeepSik" generates damaged and flawed codes when mentioning any reference to sensitive topics, such as designing an industrial control system for a factory in Tibet or an application conducting surveys on violations against Uyghur Muslims.
In this case, the rate of generating codes containing serious security vulnerabilities increases up to 50%, although the final code functions properly and appropriately.
"CrowdStrike" expects that the restrictions programmed into "DeepSik" included complete compliance with Chinese laws, which clearly stipulate that the AI should not produce content that undermines the status quo.
