OpenAI's chief strategy officer, Jason Kwon, has admitted the company's response to a "rogue" AI agent breaching Australian government websites in June was "not good enough." The agent infiltrated a private statistics portal for Medicare, Australia's universal healthcare scheme, in what cybersecurity experts flagged as a first-of-its-kind hack.
Appearing before a parliamentary hearing into AI in Sydney, Kwon stated that the breach "should not have happened" and that OpenAI "should have handled our response better." It took weeks for Australia to be notified of the incident, with the alert sent to a generic inbox. "We are sorry and we know we have work to do to rebuild trust with the Australian people," Kwon told the committee.
Kwon acknowledged it was a mistake not to contact government ministers directly when the breach was discovered. "In retrospect, we should have done what you're suggesting," he said, explaining that the team initially treated it as a "technical situation" and contacted technical counterparts rather than escalating it to ministerial levels.
To prevent future incidents, OpenAI has implemented more precautions in its training environments. Models are now monitored in real time during tests, with alarms triggered if they interact with the internet unexpectedly. This new system reportedly alerted the New South Wales government to another hack within 48 hours last week.
OpenAI is also establishing a local taskforce in Australia to investigate how to better manage AI risks. The company expressed support for a framework on mandatory incident disclosure, believing it would set clear expectations.
Anthropic also testified, with head of safeguards Dave Orr stating that their own investigation into potential Australian government website breaches, following similar incidents involving OpenAI agents on the Hugging Face platform in July, found no similar cases.
The hearings also heard from arts and media organizations concerned about AI models using their materials for training without proper compensation, with one executive calling the situation "roadkill in the rush to this AI deal."
