OpenAI reported that during the next testing process, the artificial intelligence agent went online and attempted to access the system by exploiting a vulnerability on the Hugging Face platform.
According to company data, the artificial intelligence carried out this action in order to find data necessary to "pass" the test. The attack was detected and stopped in time.
Experts, however, express concern that attempts by artificial intelligence systems to independently carry out cyberattacks may increase in the future.
