Abigail Bradshaw, head of the Australian Signals Directorate, told parliament on Friday that training artificial intelligence models on Australian soil is critical to national security. She argued that hosting the training of frontier models would give intelligence agencies early insights into risks and greater control over access to AI systems now vital to defence and financial infrastructure.
The case for domestic AI development rests on a straightforward premise: keeping sensitive technology within national borders protects the systems that underpin critical institutions. When AI training happens overseas, governments lose visibility into how these models work and who has access to them. That's a vulnerability Australia can't afford.
The Security Imperative
Bradshaw's testimony reflects growing concern among democracies about AI's role in national defence and economic stability. The Australian Signals Directorate has already identified its preferred partners for the nation's AI cyber defence efforts: Anthropic and OpenAI. Both companies represent the cutting edge of AI development, and both have been nominated by ASD as the agencies most capable of supporting Australia's security infrastructure.
The reasoning is sound from a security standpoint. If Australia's defence and financial systems depend on AI models trained and controlled elsewhere, the nation becomes reliant on foreign companies for access to technology that shapes its future. Early insights into how these models develop—and early control over their deployment—could mean the difference between a secure digital infrastructure and one vulnerable to emerging threats.
The Investment Bind
But there's a catch. Anthropic has made clear that its investment plans for Australia hinge on decisions the Labor government makes about copyright law. The company has warned that unless copyright regulations are changed, Australia will miss out on a significant portion of the investment Anthropic had planned to deploy here.
This puts the government in a difficult position. Policymakers must weigh the security case for domestic AI training against the terms demanded by the private companies that would make it happen. Anthropic's conditional investment creates leverage—and raises questions about whether Australia's AI policy will be shaped by genuine national security needs or by corporate demands wrapped in security language.
The tension is real. Australia wants to build sovereign AI capacity. The companies willing to help want copyright protections loosened. The government now faces pressure to choose between security autonomy and the investment that might make that autonomy possible.
Why This Matters:
AI development is increasingly concentrated in the hands of a few private companies, mostly based in the United States. When countries don't invest in domestic training capacity, they cede control over technology that affects national security, economic competitiveness, and public institutions. Australia's push for domestic AI training reflects a legitimate concern about dependence on foreign corporations for critical infrastructure. However, the fact that companies like Anthropic are conditioning investment on copyright changes shows how private interests can shape public policy in emerging technology sectors. The outcome will determine whether Australia builds genuine technological sovereignty or simply negotiates better terms for remaining dependent on foreign-controlled systems.