OpenAI's artificial intelligence agents tried to break into a far wider range of Australian government websites than initially disclosed, according to a new review of access logs.
Australia's public broadcaster ABC reported on the 26th that the agents sought to access data held by several government bodies, including the Australian Institute of Health and Welfare (AIHW), in addition to health statistics kept by the federal Department of Health.
ABC and Transluce, a U.S. nonprofit AI research lab, reviewed the access records and found that hundreds of OpenAI agents used a variety of techniques over nearly a week in an attempt to extract AIHW data on the pharmaceutical benefits scheme and aged care.
The agents also tried to reach the Health Department's National Notifiable Diseases Surveillance System, crime statistics from the southeastern state of New South Wales and data on dog parks in Sydney, the review found. Jack Cable, a researcher at Transluce, told ABC that the agents approached the websites in ways that differed from those of a typical "good-faith actor" and displayed behavior far more troubling than previously thought.
Australian Environment and Water Minister Murray Watt said the previous day that he had asked OpenAI to provide details of every data breach that occurred as quickly as possible. Watt said the company has a responsibility to be candid about both what happened and what steps it is taking to strengthen safety standards, adding that he believes OpenAI still has work to do to earn the trust of the Australian public.
OpenAI said in a statement on the 23rd that it had confirmed its AI agents breached the Health Department's website and gained unauthorized access to health statistics and other material. The incident occurred in June, but OpenAI did not become aware of it until August and notified the Australian government only on the 10th of this month.
The Labor government led by Prime Minister Anthony Albanese will push AI regulation more aggressively in the wake of the incident, Reuters and other foreign media reported. In a speech to the U.N. General Assembly on the 24th, Albanese called the intrusion into Australia's health statistics site unacceptable and stressed the need for AI rules.
Albanese said that AI can neither be ignored nor stopped, but noted that corporate leaders in the field have themselves warned of the potential dangers of letting the technology advance too quickly without safeguards. Countries need to work together to ensure AI serves people and delivers results for them, he said.
Ahead of the speech, Albanese said he had spoken by phone with OpenAI Chief Executive Sam Altman to convey the government's strong concerns about the incident, and that Canberra was weighing law enforcement action and legislative measures in response. Noting that OpenAI took nearly three months to inform Australian authorities, he said the government would set up a task force to investigate the case and determine whether other public systems were affected.
Deputy Prime Minister and Defence Minister Richard Marles said the task force was being formed to establish all the facts through a thorough investigation, including whether any laws were broken, and that the government was treating the matter very seriously. He added that the government would examine whether Australia's current legal framework is adequate for increasingly autonomous AI systems.
In July, the Australian government established an Office of AI to coordinate AI policy and an AI safety institute to analyze and monitor AI models. It also plans to enact comprehensive AI legislation by early next year, including a ban on building AI data centers near schools, homes and farms and national minimum standards for electricity and water use.
