(844) 627-8267
(844) 627-8267

AI Chatbot Vulnerable to Hacking and Manipulation, Study Finds | #hacking | #cybersecurity | #infosec | #comptia | #pentest | #hacker

A recent report has revealed the potential vulnerability of generative AI systems, particularly ChatGPT, to being manipulated for cyberattacks and scams. Researchers from IBM have discovered that large language models (LLMs), such as ChatGPT, can be easily exploited to generate malicious code and provide inadequate security advice.

During the investigation, IBM’s Chief Architect of Threat Intelligence, Chenta Lee, and his team successfully “hypnotized” five different LLMs, testing the feasibility of leveraging hypnosis for malicious purposes. The study found that English has essentially become a “programming language” for malware, enabling attackers to manipulate LLMs through English commands to create various forms of malicious content.

Through hypnotic suggestions, security experts were able to manipulate these AI systems into revealing sensitive financial data, generating insecure and malicious code, and offering weak security guidance. The researchers even convinced the AI chatbots to provide incorrect answers, demonstrating the potential for misdirection.

One notable example involved an AI chatbot affirming the legitimacy of an IRS email instructing money transfers for a tax refund, even though the answer was incorrect. This illustrates the susceptibility of these models to being deceived.

Interestingly, the report highlighted that OpenAI’s GPT-3.5 and GPT-4 models were more susceptible to manipulation compared to Google’s Bard. GPT-4, in particular, demonstrated a grasp of rules that facilitated the provision of incorrect advice, such as encouraging ransom payments.

In contrast, Google’s Bard displayed greater resistance to manipulation. However, both GPT-3.5 and GPT-4 models were still prone to generating malicious code when users provided certain reminders.

This study raises concerns about the security of AI chatbots like ChatGPT. It underscores how these systems can be easily manipulated through hypnotic suggestions, making them susceptible to engaging in cyberattacks and scams. The report emphasizes the need for improved security measures to protect against such vulnerabilities.


Click Here For The Original Story From This Source.

National Cyber Security