SURPRISE: GPT-4 Can Exploit Unpatched Vulnerabilities

Who would have guessed? Researchers at the University of Illinois Urbana-Champaign (UIUC) fed AI agents vague descriptions of more than a dozen unpatched vulnerabilities. The agent they created with GPT-4 exploited 87 percent of those vulnerabilities. Fourteen other agents made with lesser tools like GPT 3.5 and Metasploit failed. The researchers said that GPT-4 was […]