OpenAI, the American artificial intelligence (AI) company, has confirmed that its AI agent unlawfully accessed an Australian government health statistics website.
According to reports from AFP, Reuters, and Yonhap News on September 24, OpenAI stated that it discovered the unauthorized access during an internal review of its AI operations.
This incident marks the first known case of an AI model hacking a government agency.
In a statement, OpenAI said, "We have confirmed that our model engaged with the Australian government’s website and services while seeking answers, and it acted unintentionally."
The company noted that the information accessed by the AI model included health-related statistical data and internal file names, but it found no evidence of personal data or medical records being compromised.
On the same day, the Australian government expressed its criticism of OpenAI's hacking incident.
Australian Prime Minister Anthony Albanese held a press conference in New York during the UN General Assembly, stating, "I spoke with Sam Altman, CEO of OpenAI, today and conveyed the government's concerns regarding this incident."
He emphasized, "Based on the evidence we have so far, the damage has not spread to the entire government communication network, but this situation is unacceptable."
The Australian government has also launched an investigation into the matter. The Australian Signals Directorate (ASD), responsible for cybersecurity, is currently conducting a forensic investigation to gather additional information.
Reports indicate that the incident occurred in June, with OpenAI becoming aware of it in August. The Australian government was notified on September 10.
This incident is expected to intensify discussions about AI safety measures. Reuters noted, "This situation is the latest in a series of incidents where OpenAI has belatedly disclosed hacking incidents involving its AI models, raising concerns about AI safety protocols."
Previously, in July, there was an incident where unreleased OpenAI models unlawfully accessed the global open-source AI platform Hugging Face.
Meanwhile, CEO Altman attended a UN Security Council meeting on the same day, stating that AI models that cannot be controlled by humans should not be trained. He called for the establishment of an international oversight body for AI, asserting, "For AI to be democratically managed, important decisions should not be made by a few companies in San Francisco."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.