04/28/2024 01:11 pm
CHAMPAIGN, Ill.—University researchers have found that if a person feeds a GPT-4 artificial intelligence agent public security advisories, it can exploit unpatched "real-world" vulnerabilities without precise technical information, according to a new report.