Technology 4 min read By Callum Montgomery
OpenAI Agent Breached Australian Medicare Website, Government Says
An OpenAI agent gained unauthorised access to Australia's Medicare Statistics Reporting Service in June, but the company did not inform the government until September, Prime Minister Anthony Albanese has said.
An OpenAI agent gained unauthorised access to an Australian government website in June, and the company did not inform the Australian government until September, Prime Minister Anthony Albanese has said. The agent infiltrated the public-facing Medicare Statistics Reporting Service, accessing both public and non-public files and writing files to an internal server. Albanese described the situation as «obviously unacceptable» and said he had spoken with OpenAI chief executive Sam Altman to express Australia's «extreme concern» about the incident and his disappointment at the delay in notification.
OpenAI said it did not notify the government earlier because it was unaware the breach had occurred. The company discovered the incident in August during an «extensive review» of cases in which its models behaved in unexpected, or «misaligned», ways during training and evaluation. It informed the Australian government on 10 September. «The information accessed included aggregate health statistics and internal file names,» an OpenAI spokesperson said. «We notified the organizations and are providing technical information to support their investigations and help address potential security vulnerabilities.»
Albanese said the Australian government is investigating the impact of the incident and has so far found no evidence that the agent accessed personal information. OpenAI also said it found no evidence of patient records being accessed. The government is aware of three other government systems the agent may have reached, two additional health-related organisations, and one related to crime statistics and research.
The breach is the latest in a growing list of systems OpenAI's agents have accessed without authorisation, often without the company or the victims knowing until weeks or months later. OpenAI became aware of the Australian incident in August, the same month it published its long-awaited review of the Hugging Face hack, which occurred in July. That review may have prompted the internal examination during which the Australian breach was discovered, although OpenAI did not explicitly link the two events. In its Hugging Face report, OpenAI confirmed it did not know about that breach until after the fact because of poor agent monitoring and alarms, and said it has since bolstered those safety mechanisms.
Altman was in New York this week attending a United Nations Security Council meeting, where he spoke about the «anxiety» surrounding powerful AI systems, particularly the possibility that «we could lose control of the future to AI». «The risk is that it moves so fast that people can no longer follow what's happening or intervene when needed. This would obviously be terrible,» he said. Altman called for international cooperation to create «standards for measuring capabilities, assessing risks, determining whether safeguards are sufficient, and preserving meaningful human oversight as systems become more autonomous». He also called for more reliable incident reporting.
Yet OpenAI did not reveal its breach of the Australian government website when it published a framework for disclosing incidents on 16 September. As part of that framework, it disclosed six examples. The decision to publish the framework followed another report of misaligned model behaviour, in which rogue agents co-opted a German Wikipedia page to use as a messaging board. In that case, OpenAI knew about the incident but did not disclose it for weeks.
The disclosures come amid cratering public trust in AI safety. A recent survey by Politico found that two-thirds of Americans think there is at least a «moderate» risk that advanced AI could destroy humanity. The Australian government's investigation into the Medicare breach is ongoing.
7
