OpenAI admitted to making an error in its response after artificial intelligence agents accessed Australian government websites without permission during internal testing. The company took three months to notify the authorities and acknowledged that it should have contacted the responsible parties directly.
This incident was addressed during a parliamentary investigation in Australia into artificial intelligence, which also raised the debate about using copyrighted material in model training. Representatives from OpenAI and Anthropic were questioned regarding corporate responsibility and the limits of AI advancement in the country.
Questioning OpenAI's Conduct
During the session, independent Senator David Pocock questioned Jason Kwon, OpenAI's Director of Strategy, about the choice to send a message to a department's generic address instead of seeking direct contact with the authorities. The executive admitted that the approach taken was inadequate, stating: 'In retrospect, we should have done what you are suggesting.'
According to Kwon, employees initially treated the occurrence as a technical problem and tried to dialogue with government technical teams, but he recognized that this 'was not enough.' OpenAI communicated that it has been reviewing agent training logs since November 2025, having identified activity in June, and is seeking to verify if there are other similar events. Kwon assured that if new incidents arise, public bodies will be alerted 'very, very quickly.' The company also classified the activity on Australian websites as less complex than the attack recorded on the Hugging Face platform.
The way OpenAI managed this episode fit into a broader discussion at the hearing: as AI systems acquire the ability to interact directly with services and information, companies must be accountable not only for the functioning of the models but also for incident management.
Debate on Copyright and Model Training
This concern about the limits of corporate action also arose in the debate on model training. In this context, the conflict revolves around the use of copyrighted content and who should set the conditions for such use. Jeffrey Bleich, Anthropic's special envoy and former US Ambassador to Australia, refuted the idea that the company was trying to dictate Australian copyright rules, stating: 'We never tried to dictate Australia's' copyright norms.' He clarified that Anthropic merely detailed how training operates and what the implications of current regulations would be for its operations.
Anthropic understands that Australian legislation may require a license for every piece of content used in training, a requirement that Bleich considered 'technically impossible.' Furthermore, the company expressed interest in discussing with the government ways to stimulate model training in the country, suggesting that Australia could participate in the development of advanced models if they were trained locally.
The debate expanded to issues that transcend the technical capacity of the systems. However, the proposals from AI companies met resistance from entities representing artists, media outlets, and other professionals in the creative sector. The main concern is the risk that a change in rules would shift the burden of monitoring and controlling the use of their works onto the rights holders.
Content Creators' Concerns
This concern was voiced by Annabelle Herd, CEO of the Australian Recording Industry Association (ARIA). She warned that Australian artists could be harmed if the country modified its laws to satisfy the interests of technology companies. The ABC maintained the same stance; Kate Gilchrist, head of legal and content operations for the broadcaster, argued that a forced exclusion system would compel rights holders to continuously monitor where their work is disseminated and used.
In this landscape, the Media Entertainment and Arts Alliance (MEAA), a union representing entertainment and news professionals, proposed an alternative: a remuneration model for AI companies. This suggestion involves payments to professionals whose work contributes to these systems, in addition to creating a fund dedicated to cultural, musical, and journalistic activities. The Australian parliamentary inquiry sessions continue until Friday, with legislators debating how to reconcile the development of artificial intelligence, corporate responsibility, and the safeguarding of those who produce the content consumed by these systems.