Updated ,first published
OpenAI has apologised to Australians for its AI accessing government websites, while rival Anthropic says it cannot tell whether customers have used its models to do the same.
The two companies faced the parliament’s Joint Select Committee on Artificial Intelligence in Sydney on Tuesday, after an OpenAI agent, an AI tool that can act on its own, accessed public and non-public files on a Medicare statistics portal in June.
Prime Minister Anthony Albanese revealed the incident last month. It has raised questions about whether Australians can trust AI companies around government systems, just as those companies push to train their models here.
The incident made for a tense backdrop to Tuesday’s hearing. OpenAI chief strategy officer Jason Kwon, who had flown in from the US to attend the session, was mobbed by journalists as he entered NSW parliament, where the federal committee was convened.
Kwon told the committee the company was sorry, conceding it “should have handled our response better” and had “work to do to rebuild trust with the Australian people”.
Kwon, appearing alongside Adam Cohen, OpenAI’s head of economic policy, and Peter Anstee, its national security lead for the Asia-Pacific, said OpenAI had paused training on some frontier models, its most advanced systems, while it strengthened its infrastructure, monitoring and approach to alignment, or making sure AI behaves as intended.
He said the company had committed resources to affected agencies through OpenAI’s $1 billion Daybreak cyber defence fund and announced a local taskforce of independent Australian experts that would report by the end of the year.
When the session ended, Kwon gave muted responses to questions about how the Medicare incident was able to occur, as he left with his colleagues in a tinted van waiting outside.
Anthropic, the maker of the Claude chatbot, told the committee earlier on Tuesday that most historical customer usage of its models fell under “zero data retention” arrangements. Business customers demand these deals, under which Anthropic does not store prompts or responses.
“We can’t look through it and determine if, you know, anything like this happened,” said Dave Orr, Anthropic’s head of safeguards.
He said a review of hundreds of millions of records of the company’s own model activity had found no cases of its AI accessing Australian government systems without authorisation.
Asked what obliged Anthropic to tell the government if one of its models behaved like OpenAI’s, Orr said it was “probably controlled by Anthropic internal policy right now”. The company would notify the government within days of detecting an incident, he said.
Anthropic’s local head of policy, David Masters, said the company would welcome laws like those in California, which require AI companies to report serious safety incidents.
Jeff Bleich, Anthropic’s international special envoy and a former US ambassador to Australia, urged Australia to host the training of advanced AI models “under its own laws, with its own conditions attached”.
Under questioning, he said it was technically impossible to “license the entire internet”, and that training in Australia could be found illegal under existing law without a licence from every rights holder.
Masters and his colleague Charlie Hale appeared in the room, while Bleich and Orr joined by video.
The committee, at one point, moved Anthropic’s session behind closed doors because of the sensitivity of the discussions. Media and stakeholders were ordered out of the room and the livestream was cut.
Groups representing about 128,000 Australian songwriters and more than 40,000 authors and artists gave evidence before lunch. The music rights body APRA AMCOS warned the federal government against “caving to pressure” from US tech giants, saying AI companies must pay to train on Australian work.
Jonathan Carter, the chief operating officer of APRA AMCOS, said international music publishers were alarmed by reports the government was considering changing the Copyright Act. He said the change would let AI firms train on the world’s music in Australia with fewer obligations than in the US.
“Foreign investment by AI companies here should be on Australian terms,” he said.
Australian Copyright Council chief executive Eileen Camilleri said letting developers use creative work without permission or payment posed “an existential threat” to the sector. She welcomed the government ruling out a copyright exemption that would have let AI firms train on creative work without paying.
Australian Recording Industry Association chief executive Annabelle Herd said in a statement that Australia should welcome AI investment “but not on the condition that our artists become roadkill”.
She said OpenAI and Anthropic should negotiate with rights holders rather than behind closed doors or, in Anthropic’s case, in “secret sessions of parliamentary committees that are meant to be open to the public”.
ACT independent senator David Pocock, a member of the committee, told this masthead the frontier developers had provided “no cohesive argument” for why copyright changes were needed to train models in Australia. Nor had they explained the long-term benefit to the country, “especially in the context of paying low to no tax”, he said.
Pocock said copyright experts had warned that weakening the laws would turn Australia into a “global data laundering hotspot”, and that the hearing had raised more questions than it answered.
The ABC opened the hearing, telling the committee it has no plans to replace any of its on-screen newsreaders or journalists with artificial intelligence, and asked for help getting tech companies to pay for its content.
“To be very clear, the guidance that we give to our editorial staff makes very, very clear that they are not to create replicas of people or their voices without very careful consideration,” said David Sutton, the ABC’s senior executive for policy and regulatory affairs.
The hearings continue in Sydney on Wednesday with the retail workers’ union and Palo Alto Networks, before moving to Melbourne on Thursday and Friday, where witnesses include Amazon, Telstra, Westpac, NAB and ANZ.
The Business Briefing newsletter delivers major stories, exclusive coverage and expert opinion. Sign up to get it every weekday morning.
David Swan is the technology editor for The Age and The Sydney Morning Herald. He was previously technology editor for The Australian newspaper.Connect via X or email.
Calum Jaspan is a media writer for The Sydney Morning Herald and The Age, based in Melbourne. Reach him securely on Signal @calumjaspan.10Connect via X or email.
Elias Visontay is a National Consumer Affairs Reporter at The Sydney Morning Herald and The Age.Connect via email.From our partners
About the Author
