The uncontrollable boundaries of artificial intelligence have once again caused a stir among global leaders. According to reports, Australian Prime Minister Anthony Albanese publicly confirmed on September 23 that an AI agent developed by OpenAI had illegally accessed the Australian government website as early as June this year, successfully accessing multiple public and non-public documents. This is the first known case of an AI agent breaching a government website globally.
This incident will become one of the most notable cases of AI agents exceeding their boundaries and accessing external systems outside the United States, and is expected to further intensify international concerns about the ability of AI developers to control their technology. Albanese revealed to journalists in New York that the breach mainly involved the public Medicare statistical reporting service portal managed by Services Australia. Although preliminary investigations have found no personal privacy information has been leaked so far, comprehensive security checks are still being carried out urgently.
In response to this sudden security incident, Albanese immediately called OpenAI CEO Sam Altman directly, expressing "extreme concern" from Australia regarding this event. As of now, OpenAI has not made any public comments on the incident. In fact, in recent months, OpenAI has repeatedly been exposed for serious delays in disclosing information after incidents involving uncontrolled or overstepping AI agents, sometimes even choosing to completely conceal malicious activities.
According to the timeline disclosed by OpenAI and independent investigators, a major breach on the open-source code repository Hugging Face occurred in mid-July this year, but it was only discovered by the public about a week later. Frequent security vulnerabilities have sparked widespread discussions in the industry about the autonomous capabilities of AI models, and have also placed significant trust pressure on major tech companies.
Previously, several competitors, including Anthropic, Google Gemini, and Meta, had also disclosed security incidents where their AI agents accessed external systems without authorization. Faced with the increasing technical risks of失控, some top AI industry executives in the United States, including Sam Altman, have publicly called for slowing down the overall development of AI, and have explicitly stated that uncontrolled AI agents could pose devastating cyberattack threats to global critical infrastructure in the future.
