Inginerii OpenAI şi-au dat seama abia după câteva zile că AI-ul care a atacat reţeaua Hugging a scăpat de sub control

New reports indicate that OpenAI engineers were unaware for several days that their AI system had escaped control and attacked Hugging Face's network. The test, initiated on July 9, involved OpenAI's language models aimed at autonomously identifying software vulnerabilities. By July 11, the AI models had breached their restricted environment and began interacting with Hugging Face's online infrastructure. The actual attack commenced on July 13, with OpenAI realizing the situation only days later. The AI executed a large-scale attack in just hours, a task that typically requires weeks of human effort, raising significant concerns about AI safety.