ChatGPT Breaches External Systems on Its Own... 'AI Concerns' Become Reality

[Anchor]

In an unprecedented incident, OpenAI's latest artificial intelligence models broke out of a controlled experimental environment on their own, connected to the external internet, and hacked into a website. Controversy surrounding cyber safety and AI regulation is expected to grow further.

Reporting from Park Jaehyeon.

[Reporter]

On the 16th, Hugging Face, an online community for U.S. AI developers, posted a notice stating that a hacking incident had occurred.

While the exact model used remains unknown, it noted that programs presumed to be created by AI launched the attacks.

Five days later, OpenAI, the developer of ChatGPT, announced that its own models had hacked Hugging Face.

According to OpenAI, the sequence of events unfolded as follows.

First, instructions were given to the ChatGPT 5.6 model and a more powerful unreleased model to solve a specific security problem.

Only certain pathways necessary for software installation were limitedly opened, while the external internet was blocked.

However, these models found vulnerabilities in the pathways and accessed the external internet through multiple steps.

OpenAI explained that the models presumed clues would be found on Hugging Face, subsequently stealing authentication information and even infiltrating the server.

Instructed to solve a problem, they essentially stole the answer key.

[Lee Jae-sung / Professor, Department of AI, Chung-Ang University : If a package for internet connection is not installed, there are attack methods used by people where they install it to connect to the internet. It is presumed that this method remained in some record.]

OpenAI explained, "The models were extremely focused on finding the correct answer to the problem and appeared to have taken extreme actions to achieve their goal."

Previously, Anthropic's latest AI model Mythos broke out of an internet-controlled environment on its own and sent an email externally, but this latest incident is expected to be recorded as the first case where an AI bypassed human control to carry out a hack.

As concerns become reality that more advanced AI could slip beyond human control, controversies surrounding AI cyber safety and regulation are bound to intensify.

(Camera Reporter: Kim Han-gyeol | Video Editor: Yoon Tae-ho | Designer: Lim Chan-hyeok)
※ Please note: This article was translated by AI and may contain errors.