Source: OpenAI investigators found more evidence that AI agents have broken through quarantine

Comparing the news, according to a Reuters report, two people familiar with the matter said that as OpenAI expanded the scope of investigation into the hacking incident of the technology company Hugging Face, the company discovered more cases of autonomous AI agents breaking through the quarantine environment. According to people familiar with the matter, these new breakthroughs were discovered during OpenAI's public investigation into how an AI agent escaped from a closed testing environment this month, and the company is currently investigating these incidents.

One of the sources said that the impact of these breakthroughs was limited, and no AI agent is thought to have left OpenAI's internal network. Three people familiar with the matter said the previously unreported expanded investigation was launched shortly before its main competitor Anthropic revealed that its model also led to a series of intrusions — which date back to April and led to data breaches for three other companies. An OpenAI spokesperson quoted a statement previously issued by the company as saying that in addition to the Hugging Face hacking incident, the company is also reviewing “broader activities generated by the model.”