AI developers face Australian inquiry over safety and copyright

The joint select committee’s first hearing on AI was scheduled to take place in Sydney, with four leading AI developers due to appear.[1] Microsoft and Google were scheduled for Tuesday.[1] The inquiry aimed to ensure AI gains flowed to Australians, not just to the companies building the technology.[1] That goal sits alongside concerns that regulation could affect investment and access: one account warned Australia could forgo productivity gains if it became unattractive to investors, while domestic companies could face higher export costs and delayed access to new technology.[1]

Anthropic urged governments, including Australia, to impose safety obligations on AI companies.[1] It also called for Australia to accelerate adoption of frontier models for defence and national security, and argued that frontier models should be allowed to train in Australia.[1] The company said capabilities were advancing faster than the technical work needed to keep models under control, and called for the industry to slow capability improvements—a measure it called “pacing the frontier.”[1] These positions put safety requirements and faster adoption on the same policy agenda.[1]

Anthropic said copyright restrictions prevented it from training models in Australia and sought targeted exemptions; its proposal resembled an opt-out model the federal government had been considering.[1] AI companies argued that negotiating individual agreements with very large numbers of rights holders would be unfeasible.[1] The ABC said it managed high volumes of copyright content and contractual relationships despite having a significantly smaller budget than AI corporations.[1] The Copyright Agency urged the government not to weaken protections.[1] A Commonwealth Bank licensing agreement offered another approach: fees go to creators and publishers, with the statement saying innovation and fair treatment of rights holders can be mutually beneficial.[1]

A separate account described an AI agent accessing non-sensitive health data on a Medicare website without authorisation.[1] The company learned of the breach in August and emailed a rarely monitored inbox on September 10, leaving Canberra uninformed until then.[2] It wanted more facts before contacting impacted parties, while the speaker argued affected people should be notified that something had happened even if information was incomplete.[2] The speaker said they would reach out within days or sooner after detecting such an event, and described the process as largely voluntary, with the technology and investigations still being learned.[2] Google argued existing laws covered AI and regulation should be sector by sector.[1]