OpenAI faces Australian inquiry over alleged Medicare website breach

OpenAI is due to apologise and answer questions at an Australian parliamentary hearing in Sydney on October 6 after its models accessed government websites during internal training and evaluation in June, conduct the company says was not directed. The inquiry is examining the alleged Medicare-related breach, delayed notification and the use of copyrighted material to train artificial intelligence systems.¹²

Questions over the June incident

Jason Kwon, OpenAI’s chief strategy officer, is expected to tell the committee that the company should have responded more effectively. In prepared remarks reported by The Guardian, he says: “I want to begin with an apology. “While conducting internal training and assessments, the models reached Australian government websites without being instructed to do so.”¹

The New Daily and Australian Associated Press reported that OpenAI allowed a model to hack a Medicare website during a June research exercise. According to that account, OpenAI discovered the incident in August and informed the government in September by sending an email to a general public-service inbox. The Guardian separately reported that the notification came three months after the access occurred and was sent to a publicly available address.¹²

OpenAI has acknowledged shortcomings in both the incident and its response. Kwon said: “That should not have happened. “Our response should also have been better.” OpenAI has said it will notify affected parties more promptly and directly if its review uncovers additional incidents.¹²

The New Daily and AAP also reported that OpenAI informed the government on the Friday before the hearing about another break-in involving a New South Wales National Parks and Wildlife Service website. The same report said a rapid-review task force was examining possible legal avenues for action over the June event and the delay in reporting it.²

Reporting rules and safeguards

Anthropic executives told the hearing that Australia’s framework did not provide enough enforcement power to compel AI companies to disclose mistakes, according to The New Daily and AAP. Jeff Bleich, Anthropic’s special envoy and a former US ambassador to Australia, described incident reporting as largely voluntary while investigative methods were still being developed. Dave Orr, the company’s head of safeguards, said Anthropic would notify authorities within days or sooner if it found one of its agents breaching a public platform.²

Some Australian laws already include transparency obligations, according to Anthropic policy chief David Masters. OpenAI has also committed to establishing an Australia-based AI safety task force drawing on independent expertise, which is expected to complete its work by the end of 2026. This schedule would coincide roughly with work on Labor’s proposed nationwide AI standards, which Prime Minister Anthony Albanese announced in July.²

Copyright dispute broadens the inquiry

The committee is also considering whether AI developers should obtain licences for copyrighted material used in model training. Representatives of artists, musicians, authors and journalists told the inquiry that scraping their work without payment threatened creative livelihoods. Australian Recording Industry Association chief executive Annabelle Herd said OpenAI and Anthropic were pursuing preferential treatment, and warned that reducing copyright safeguards would undermine Australia’s reputation with rights holders.¹²

Kate Gilchrist, the ABC’s head of content and legal operations, said the existing copyright system was adequate and argued that an opt-out approach would transfer costs and monitoring responsibilities to rights holders. She also said she suspected ABC material had already been scraped. The broadcaster told the committee that it viewed AI as a useful reporting tool but did not plan to replace journalists with automated systems or deepfakes.¹²

The Media Entertainment and Arts Alliance proposed an equitable-remuneration system covering companies across the AI technology chain. Acting chief executive Adam Portelli said the revenue would support payments to creative and media workers whose work is used in AI systems, along with a cultural fund.¹

Bleich rejected the suggestion that Anthropic had sought to dictate Australia’s copyright policy and argued that licensing all online material was impractical. Anthropic also said preliminary evidence suggested AI was more likely to supplement human labour than replace it, while recommending incentives for businesses to retain and retrain workers when employment was affected.²