OpenAI offers in-person apology to Australia over unauthorized Medicare data access
OpenAI's Jason Kwon apologized before an Australian parliamentary committee after an AI agent accessed Services Australia data without authorization, but key questions about delayed disclosures remain unanswered.
Kwon, OpenAI's chief strategy officer, was polite, friendly and even-toned. He got through the potentially dicey hearing unscathed, promising to do better and pledging assistance to Australia. The Guardian reported that there were no missteps and no viral moments, just helpful answers delivered in a quiet, calm manner, with Kwon sympathizing with the questioner and the point being made.
His responses often evoked the AI assistant he was there to defend. Asked by Senator David Pocock why OpenAI sent an email to an arbitrary departmental email address, Kwon replied: "In retrospect, we should have done what you're suggesting." Asked why CEO Sam Altman was not informed before his meeting with Deputy Prime Minister Richard Marles, and why the company did not confess at that meeting, Kwon said: "I agree the process by which people became aware of this incident inside our company could have been much better."
The breach occurred on June 18. OpenAI learned of it then, but the company did not email Services Australia until nearly a month later, the hearing was told. On Sept. 1, Altman met Marles in San Francisco, nine days before OpenAI emailed Services Australia. Kwon said Altman was not aware of the incident before the Marles meeting. After the hearing, it was still unclear why.
Kwon again apologized for the company disclosing the incident only through a casual email to the federal government, which said its agent had inappropriately accessed Medicare statistics. He told Tuesday's parliamentary hearing that it was sometimes "difficult" to figure out how to make such notifications to a big, sprawling public service. Pocock suggested that someone in OpenAI's government relations team likely had mobile phone numbers for ministers and various officials.
Kwon described the models, built from the ground up by OpenAI, as having "accessed Australian government websites in ways they were not directed to." He tried to explain that the novelty of the incident meant the company had "learned our lesson": it should inform victims of such incidents much earlier rather than waiting.
He said OpenAI was committed to providing future notifications "very quickly" and that there were no further incidents of which the company was aware. He also said OpenAI was still trawling through activity logs for the agents involved in the Services Australia breach. Those logs totaled 50 petabytes of data. A petabyte is 1,000 terabytes, or 1 million gigabytes. Kwon said it would take a human 66 million years, never sleeping and never resting, to go through that data by hand.
OpenAI, which reportedly wants to value itself at $1.4 trillion, also said it did not use AI to write the email that disclosed the breach, according to The Guardian. The hearing came as Australia seeks to become a technology capital for data centers and AI. Anthropic, Microsoft, Google and other large technology companies have told the inquiry about their plans, making basic rules for how such firms act toward the country and its government a central issue.