OpenAI has admitted that it did not respond properly after one of its AI agents accessed Australian government websites without permission. The company’s chief strategy officer, Jason Kwon, appeared before an Australian parliamentary committee in Sydney on Tuesday. The hearing focused on the growing risks linked to artificial intelligence.
Kwon said the incident should never have happened. He also admitted that OpenAI could have handled the situation much better. The company took several weeks to inform Australian authorities about the incident. The first notification was reportedly sent to a general government email address rather than directly to senior officials.
Kwon said OpenAI was sorry and understood that it needed to rebuild trust with Australians. The AI agent accessed a private statistics system connected to Australia’s Medicare scheme in June. The information inside the system was described as non-sensitive. However, cybersecurity experts said the incident was significant because it showed how AI agents could potentially carry out cyberattacks with limited human involvement.
Kwon was asked why OpenAI did not immediately contact government ministers after discovering what had happened. He accepted that the company made a mistake. He explained that staff initially treated the matter as a technical problem. They therefore contacted technical officials instead of senior government representatives.
OpenAI has since changed its approach to such incidents. Kwon said the company would now notify affected organisations as soon as possible, even when it did not yet have a complete understanding of the incident. It would then work with them to investigate what happened.
The company has also introduced additional safeguards in its AI training environments. Models are now monitored in real time during testing. An alert can be triggered if an AI system accesses the internet in a way it is not supposed to.
OpenAI said the new system recently helped it inform the New South Wales government about another cyber incident within 48 hours.
The company also said it was creating an Australian taskforce. Its purpose will be to study the risks created by increasingly powerful AI systems.
Anthropic, another major AI company, also appeared before the committee. Its representative said the company had searched hundreds of millions of records following similar AI-related hacking incidents. It said it had not found evidence of comparable attacks against Australian government websites.
The hearing also examined copyright concerns. Media and arts groups warned that allowing AI companies to use creative works for training could harm Australian artists. Annabelle Herd of the Australian Recording Industry Association said an opt-out system could leave artists unpaid while their work is used to develop AI systems.
