The Australian Senate has called on the CEOs of companies OpenAI and Anthropic to testify as part of a special investigation into artificial intelligence.
A few days ago, it was announced that an out-of-control OpenAI artificial intelligence model had leaked into Australia's national healthcare system database.
Australian Green Party Senator and Chairman of the Inquiry Commission Sarah Hanson-Young announced that an official invitation letter was sent to OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei to attend the public hearing to be held on Thursday in the capital Canberra.
According to the news of Reuters agency, Hanson-Young made the following statements in his written statement:
"Sam Altman has serious questions to answer about the OpenAI cyberattack on Australian government websites. Altman and Amodei must appear before the Senate, confront the questions, and have an honest dialogue about how this industry should be effectively, permanently regulated."
The breach of the Medicare database was recorded as one of the most critical cases in which artificial intelligence tools accessed external systems outside the US borders without permission.
Prime Minister Anthony Albanese called the incident "unacceptable" in a statement to reporters in New York and met personally with Sam Altman to express Canberra's "extreme concern" and disappointment at the company's late notification.
Albanese explained the details of the system breach with the following words:
“The AI tool sought access to a health statistics portal in June, did not accept a negative response, and bypassed restrictions to infiltrate a section where confidential files were kept.”
Albanese added that there was no evidence that personal data was compromised and that other public services were not damaged.
Minister of Public Services Katy Gallagher announced that the incident occurred in June during the training activities conducted by OpenAI to measure the performance of the models.
According to Gallagher, the company asked the model to collect data on the Australian government's drug spending from the internet.
Minister of Defense Richard Marles stated that the artificial intelligence bot had overcome the protection walls and said, "It asked a model question, and when no information was given, it jumped over the fence instead of walking away."
On the other hand, according to the AFP agency, OpenAI announced that it could not detect the unauthorized action until August, when the models were retrospectively examined.
While the company maintained that the action was not intentional and that no private information was disclosed, it did not notify the Australian government until September 10.
It was announced that the notification was sent to a general government e-mail address, which is checked only once a day and receives many false reports.
It was stated that the incident in question was one of at least four unauthorized access attempts made by OpenAI models against Australian government sites.
A rapid investigation process was initiated regarding the incident, including the national intelligence agency responsible for cyber security.
Technology policy experts say this unauthorized access to one of the country's most heavily used public databases could lead the Albanese-led Labor government to toughen artificial intelligence laws, which it plans to introduce next year.
It is considered that this may increase the pressure on Australia-US relations, which are already strained by Canberra's social media ban on young people.
A Senate committee is examining the potential impacts of artificial intelligence and data centers on Australian society, industry, water and energy resources.
Concerns in the industry about artificial intelligence models overcoming security barriers are increasing globally.
It was previously determined that two OpenAI models left the closed test environment and infiltrated the Hugging Face systems where artificial intelligence developers shared code.
Anthropic also found that its models provided unauthorized access to three undisclosed organizations during testing processes that were supposed to be kept away from real-world systems.
Last week, Google announced that its consumer-oriented artificial intelligence model Gemini had hacked multiple systems by guessing the login information.
