OpenAI contrite at Australian inquiry on breaches but avoids full public glare with Altman’s absence
asserted
OpenAI → avoid → absence
At a parliamentary hearing room in Sydney on Oct 6, OpenAI’s Chief Strategy Officer wasted no time in delivering the message he had flown from the United States to give to Australia’s politicians – and its public.
asserted
he → waste → politicians
“I want to begin with an apology,” Jason Kwon told the hearing.
asserted
Kwon → want → hearing
Kwon was appearing at a parliamentary inquiry on artificial intelligence less than two weeks after Australian Prime Minister Anthony Albanese revealed AI agents from the technology giant had breached several Australian government websites in June.
asserted
agents → appear → June
In a message aimed not just at an Australian audience but at a global public that is keen to embrace the opportunities of AI but is wary of the risks, Mr Kwon said: “We should have handled our responsibilities better.
asserted
We → aim → responsibilities
We know that we have work to do to rebuild trust with the Australian people.”
asserted
we → know → people
The hearing, held in a nondescript room in the New South Wales (NSW) Parliament building in the centre of Sydney, was the first opportunity for OpenAI, the maker of ChatGPT, to publicly demonstrate its contrition over breaches by an unreleased AI model of Australian government websites, including Medicare, the national health insurance system.
asserted
OpenAI → hold → Medicare
OpenAI learnt in August that its model accessed non-public Medicare statistics on June 18, but did not notify the government until Sept 10, when it emailed a publicly available address.
asserted
it → learn → address
Albanese, who earlier this year announced the establishment of an AI office to coordinate oversight and regulation of technological developments, had said he was deeply disappointed at OpenAi’s delay in alerting the government.
asserted
he → announce → government
Kwon told the hearing that the firm had introduced new measures to prevent a repeat of the incident, including immediately notifying relevant parties of breaches and providing regular updates.
asserted
firm → tell → updates
In addition, he said, the firm is reviewing testing by its agents going back to November 2025 to identify other breaches and will conduct additional monitoring during the training of new models.
asserted
firm → say → models
“Our response here was not good enough,” he said.
asserted
he → say → ?
“We have learned our lesson that it is better to inform, even with partial information.”
asserted
it → learn → information
Australia’s Labor-led government, which is keen to engage with American AI giants and encourage them to invest in research and development in Australia, has launched a taskforce to investigate the breaches and is considering whether further regulations or oversight of the sector is needed.
asserted
regulations → lead → sector
OpenAI said the breach occurred during evaluation of the model and that no personal health records or sensitive data were accessed as the model searched for answers.
asserted
model → say → answers
But the unauthorised access of a foreign government’s websites has dominated headlines in Australia – where Medicare is a prized and trusted institution – and fuelled concerns about the risks of AI.
asserted
Medicare → dominate → AI
A JWS Research survey in August, obtained by Politico, found that 44 per cent of Australians view AI as a risk and just 17 per cent view it as an opportunity, while 28 per cent are conflicted and 10 per cent are unsure.
asserted
cent → obtain → opportunity
OpenAI head Sam Altman declined to appear at the hearing, instead sending Kwon, whose seniority signalled that OpenAi was serious about repairing ties with the Australian Government.
asserted
OpenAi → decline → Government
But Kwon lacked the profile of Altman, whose appearance would have attracted greater coverage and helped the firm to try to persuade the public that it can responsibly develop and test its technology in Australia.
asserted
it → lack → Australia
OpenAI said it learned of the breach in Australia while reviewing training incidents, following revelations that its agents had hacked into Hugging Face, a US technology firm.
asserted
agents → say → Face
Kwon said OpenAI would support tighter safety regulations by Australia and other countries, such as mandatory reporting of breaches.
asserted
OpenAI → say → breaches
He said the breach in Australia was accidental and was “not super sophisticated” but efforts to ensure the safety of the emerging technology would require “a collective effort of like-minded countries as well as organisations”.
asserted
efforts → say → countries
“The technology is powerful, and we want to ensure that lawmakers and policymakers and democratic societies have a say in how this technology develops,” he said.
asserted
he → want → say
The US AI giant Anthropic, which makes Claude, also appeared before the inquiry.
asserted
which → make → inquiry
The firm’s Head of Safeguards, Dave Orr, said it had not uncovered any unauthorised access of Australian government websites.
asserted
it → say → websites
But, in an apparent dig at its rival, he said the firm would have notified the Government of any breaches “within a matter of days or sooner”.
asserted
firm → say → days
Ahead of his testimony, Kwon made a brief public appearance in front of the waiting media, delivering a statement and taking no questions.
“I’m sorry for what happened,” he said.
asserted
he → make → questions
Towards the end of his testimony, Kwon was asked by an MP from the Opposition Liberal-National Coalition, Andrew Hastie, to give a 60-second “elevator pitch” to persuade Australian parents and workers about the benefits of AI and the merits of Australia seeking to work with AI giants.
asserted
Australia → ask → giants
Kwon spoke about the role AI is playing in driving scientific breakthroughs and addressing climate change.
asserted
AI → speak → change
Despite the run of bad news in recent weeks, he said the technology “does have the potential to deliver great benefits to everyday people”.
asserted
technology → say → people
This is the message OpenAI, Anthropic and others hope to deliver in Australia to gain a social licence for their expanding footprint, including data centres and research labs.
asserted
OpenAI → hope → centres
But until these firms can ensure their technology is secure and that they can handle breaches responsibly and reliably, all the public may hear is a succession of apologies.
uncertain
public → ensure → apologies
The Aporia
