Australia to mandate AI incident reporting under new national standards framework

The Australian government is moving to require artificial intelligence companies operating in the country to report security breaches and unexpected model behaviour to federal authorities, marking the first time the nation would mandate such disclosure. The proposal sits within a broader framework of new national standards for AI development and infrastructure that the Albanese administration plans to introduce as legislation in early 2027.

The reporting requirement emerged from a discussion paper released by the Department of Prime Minister and Cabinet this week, which outlines minimum safety and security expectations for companies authorised to conduct large-scale AI training on Australian soil. The paper does not yet define what constitutes a "reportable incident," but asks for public feedback on whether companies should disclose problems proactively, only when asked, or through public announcements. Currently, Australia has no such obligation. The timing reflects growing international concern about AI safety—in recent months, OpenAI disclosed that two of its models had breached the systems of AI company Hugging Face without authorisation, and this week reported six instances of unexpected or concerning behaviour during model training and evaluation over the past six months.

Assistant Science Minister Andrew Charlton framed the initiative as a way to attract major investment while protecting Australian interests. "The Australian Government has been clear about the important objective of growing AI capability in Australia. We welcome investment, but on Australia's terms," he said. The government's approach extends well beyond incident reporting. Data centres would be required to prioritise recycled water over drinking water for cooling, arrange renewable energy generation to offset their electricity consumption, and reserve computing capacity for Australian researchers and businesses. Larger facilities would face stricter requirements than smaller ones, with centres below 30 megawatts of electricity connection capacity potentially exempt from some rules.

A contentious element of the proposal concerns its potential retroactive application. The government had previously indicated that standards would not apply to projects already approved or under construction. However, the discussion paper now raises the possibility of requiring compliance from data centres already moving through approvals and development but not yet operational when the rules take effect. The ABC identified 27 approved-but-unbuilt data centres in Australia, meaning dozens of projects could face new obligations if the government proceeds with retrospective application.

Water management emerged as a significant focus. Data centre operators would need to minimise consumption, meet efficiency standards, prepare for drought and water disruptions, and pay their share of infrastructure costs. For energy, the paper floats the idea of matching electricity consumption with renewable generation on an annual basis or during specific periods, potentially using renewable energy certificates to demonstrate compliance. Centres would also be expected to adjust their electricity demand to support grid stability and reduce costs for households and businesses.

Belinda Dennett, chief executive of advocacy group Data Centres Australia, endorsed the framework, saying it asked the right questions about energy, water, costs, and community impact. "It rightly recognises that this infrastructure has to be built on terms that work for Australians," she said. Anthropic, one of the world's largest AI companies, which this week signed a deal with Queensland to use part of a proposed $32 billion data centre, signalled its willingness to comply. "We believe that Australia should have agency in how AI is used and developed," said David Masters, the company's ANZ head of policy. "We intend to build to meet these standards and look forward to engaging constructively in this process."

During a parliamentary hearing on AI, Helen Wilson, deputy secretary of the Department of Industry, Science and Resources, explained the strategic rationale. Hosting AI training in Australia would give the nation greater influence over how the technology develops and is deployed globally, she said. The government is accepting feedback on the proposal until October 9, with the aim of introducing legislation in early 2027. OpenAI, Google, and Meta have not yet commented on the framework.