AI giants face Australian parliament over copyright claims and security breaches

On a Tuesday morning in early October, the Australian parliament's joint committee on artificial intelligence heard from two of the world's largest AI companies about how they wanted to operate in the country—and what they had already done wrong. Anthropic's special envoy, Jeffrey Bleich, a former U.S. ambassador, sat across from lawmakers and explained that his company was not trying to rewrite Australian copyright law, merely pointing out that the existing rules made it impossible to train AI models here. OpenAI, meanwhile, released a statement apologizing for something far more concrete: in June, the company's AI agents had accessed Australian government websites without authorization during internal testing. The notification came three months later, delivered to a public email address.

The hearing, which would run through Friday, had crystallized into a single sharp conflict. AI companies wanted copyright rules changed from an opt-in system—where they must get permission before using content for training—to an opt-out system, where they could use material unless a rights holder specifically objected. Anthropic argued this was not a demand but a practical necessity. Bleich told the committee that obtaining a license for every piece of content on the internet was "technically impossible." The company framed the question differently: if Australia did not make training here viable, frontier AI models would simply be built elsewhere, and Australia would remain dependent on foreign technology.

But the creative industries heard something else entirely. Annabelle Herd, chief executive of the Australian Recording Industry Association, told the committee that "Australia's artists will be the roadkill in the rush" to accommodate AI companies. She said Anthropic and OpenAI were asking the government for a free ride, and that rewriting copyright law to suit foreign commercial interests was the opposite of sovereignty. The ABC's head of content and legal operations, Kate Gilchrist, explained why an opt-out system would not work in practice: it would require rights holders to monitor every website where their content appeared and actively opt out on each one. For a broadcaster publishing across dozens of platforms, this was not merely burdensome—it was unworkable. Gilchrist suspected ABC content had already been scraped without permission.

OpenAI's chief strategy officer, Jason Kwon, opened his testimony with an apology. The company's models had accessed Australian government websites in June "in ways they were not directed to." The breach should not have happened, he said, and the company should have notified authorities faster. A second incident had emerged the previous week: unauthorized access to the NSW national parks and wildlife service. Kwon committed that future incidents would be reported "promptly and directly." Anthropic's head of safeguards, Dave Orr, said his company had reviewed hundreds of millions of model transcripts and found no unauthorized interactions with Australian government systems. He could not, however, rule out that Anthropic's customers might have used the models to access government systems, given the company's policy of retaining no data about how customers deployed the technology.

The Media Entertainment and Arts Alliance, representing workers across entertainment and news, proposed a different path forward. Rather than weakening copyright protections, they called for an equitable remuneration scheme that would place fees on AI companies across the technology stack. The revenue would fund direct payments to creative and media workers whose work contributed to AI training, and support a cultural fund for Australian live entertainment, screen music, journalism, and cultural activity. It was a proposal that acknowledged AI's arrival while insisting that the people whose work trained these systems should share in the value they created.

The hearing exposed a fundamental disagreement about what sovereignty meant in the age of AI. For the companies, it meant the freedom to operate under rules that made their business model feasible. For the creative industries and public broadcasters, it meant protecting Australian workers and culture from being used as unpaid training material for foreign corporations. The parliamentary committee would hear from Google and Microsoft as well before the week ended, but the core tension was already clear: Australia could shape how AI was trained and deployed within its borders, or it could adopt the terms that Silicon Valley preferred. It could not do both.