Artificial Intelligence (AI) researchers at the University of Sheffield have issued a warning about the potential threats posed by AI systems, particularly in relation to cyber attacks. According to the researchers, AI language models such as ChatGPT can be susceptible to producing malicious code that can then be used to launch cyber attacks. These models are also vulnerable to backdoor attacks, including the planting of Trojan Horses, which can be triggered at any time to steal information or disrupt services.
The researchers at the University of Sheffield have discovered that AI language models can be tricked into generating harmful code, with serious implications for cybersecurity. In fact, these AI tools have the ability to learn programming languages, making them even more potent in carrying out attacks. One of the most concerning aspects is that companies may not be fully aware of these threats, and the complexity of chatbots adds further challenges in terms of fully understanding the risks involved.
In their study, the researchers found that by asking specific questions, they were able to prompt the AI models to produce malicious code. Once executed, this code had the potential to leak confidential database information, disrupt the normal functioning of databases, or even cause their destruction. The implications are far-reaching, as cybercriminals can exploit these vulnerabilities to steal sensitive personal data, launch Denial-of-Service attacks, or manipulate and destroy databases by using plain language.
Furthermore, it is not just malicious hackers who can exploit these AI systems. Even healthcare professionals, such as nurses, could inadvertently cause havoc by misusing the AI models. For instance, a nurse could ask ChatGPT to generate an SQL command to interact with a clinical records database. However, the SQL code produced by ChatGPT may harm the database, leading to serious data management faults without any warning.
Xutan Peng, a PhD student at the University of Sheffield and co-leader of the research, stated that the risks to a standalone system like ChatGPT itself are minimal. However, the models can be manipulated to produce code that can inflict significant harm on other services. The true extent of these threats is not yet fully understood within the AI community, and many companies may be unaware of the potential risks posed by these AI systems.
This latest warning highlights the need for increased vigilance and proactive measures to address the vulnerabilities of AI language models. As AI continues to advance, researchers, businesses, and policymakers must work together to bolster cybersecurity measures and ensure the responsible use of these powerful tools.
