OpenAI’s GPT-4 exploits 87% daily vulnerabilities autonomously. #cybersecurity

A new study by University of Illinois Urbana-Champaign researchers found that the GPT-4 large language model from OpenAI can autonomously exploit real-world vulnerabilities, specifically one-day vulnerabilities that have not been patched yet. This capability sets GPT-4 apart from other models and vulnerability scanners. The researchers noted that GPT-4’s success in exploiting vulnerabilities is due to its advanced tool use capabilities.

The study revealed that GPT-4 was able to exploit 87% of one-day vulnerabilities, showcasing its emergent capability in autonomously detecting and exploiting vulnerabilities. The cost of using GPT-4 to exploit vulnerabilities was found to be significantly lower than employing a human hacker, making it a cost-effective option for cybercriminals.

The experiment involved testing GPT-4’s vulnerability exploitation capabilities by providing it with detailed prompts and assessing its success rate in exploiting vulnerabilities. The researchers compared GPT-4’s performance with other LLMs and vulnerability scanners, finding that only GPT-4 could successfully exploit one-day vulnerabilities.

Overall, the study aimed to address the gap in knowledge regarding LLMs’ ability to exploit vulnerabilities without human intervention. The researchers demonstrated the potential of using advanced AI models like GPT-4 for automating the process of exploiting vulnerabilities in computer systems.

Source link

Source link: https://www.techrepublic.com/article/openai-gpt4-exploit-vulnerabilities/