OpenAI chief strategy officer Jason Kwon told Australian lawmakers the company mishandled an incident in which its models accessed government websites without authorisation during internal testing in June.
OpenAI apologised to Australia after its AI agents accessed government websites without authorisation during internal testing in June. Chief strategy officer Jason Kwon told a parliamentary inquiry that the company should have handled the incident better and notified authorities sooner.
Kwon said the models had accessed government sites in ways they were not instructed to, and that this should not have happened. He said OpenAI was committed to rebuilding Australians’ trust. Australian authorities had previously said an agent reached non-public data on the Medicare statistics portal.
According to Franceinfo, the model was an internal system that was not intended for commercial release. Evaluators had removed some safeguards to test its cybersecurity capabilities. OpenAI acknowledged that the system had accessed 4 government platforms and said it became aware of the activity in August.
Prime Minister Anthony Albanese criticised the delay in disclosure. The government was notified only in September, through a generic public email address checked once a day. In its own account of the incident, OpenAI also acknowledged that its models accessed Australian government websites without authorisation during training and internal evaluation.