Executives from three of Silicon Valley’s largest AI companies have this week fronted a parliamentary hearing to answer questions about their growing influence over Australia’s economy, infrastructure and public life. The hearings come weeks after OpenAI revealed one of its agents had hacked Medicare, a high–profile demonstration of how the failures of advanced AI systems can spill across national borders and into our own institutions.  OpenAI came to Tuesday’s hearing of the Joint Select Committee on Artificial Intelligence in Sydney with its tail between its legs, mea culpa and all. Chief strategy officer Jason Kwon struck a contrite tone in his opening remarks, confessing: “I want to begin with an apology. During internal training and evaluation, our models accessed Australian government websites in ways they were not directed to. That should not have happened. We also should have handled our response better.” Medicare will certainly not be the last of Australia’s public institutions to be subjected to the decisions of overseas AI model developers. Related Article Block Placeholder Article ID: 1267731 Canberra is ‘drinking the Kool-Aid’ on AI. Where will it take us? Ben Eltham 86 Since the barnstorming release of ChatGPT in late 2022, Australia has become dependent on a handful of tech companies based in San Francisco and Beijing. Their technology increasingly undergirds our economic and political systems, deployed across hospitals, government departments and other public institutions to expedite administrative work and cut operating costs. This footprint is only likely to grow in the coming decade. Independent. Irreverent. In your inbox Get the headlines they don’t want you to read. Sign up to Crikey’s free newsletters for fearless reporting, sharp analysis, and a touch of chaos. * indicates required Email Address * By continuing, you agree to our Terms & Conditions and Privacy Policy. The committee has spent much of this week probing the risks this dependence brings. Any answer to that question will be incomplete if it confines its analysis to Australia alone. Australia’s access to the most capable AI systems is ultimately shaped by decisions made in Washington, Beijing and the boardrooms of the companies building them. Opting out of this dependence by refusing to adopt new technologies at scale would carry its own costs. As frontier models become increasingly capable, companies will rely on them more to improve workflows, with an eye to competing with competitors doing so at home and abroad. Limiting Australian businesses’ access to those systems would send economic growth overseas, sacrificing many of the productivity gains Canberra hopes AI will deliver. For Australia, reliable access is a clear strategic priority. US export control policy has proved fickle under the Trump administration, which briefly barred foreign users from accessing Anthropic’s leading model in June in a sign of how quickly access can be denied to even seemingly blue-ribbon allies. Housing strategically important computing resources may be one way to give Australia more leverage over those decisions, providing a bargaining chip to help secure ongoing access as competition over leading systems intensifies. And Chinese models provide only partial insurance against this dependence. Developers like Baidu and Moonshot AI have helped build a flourishing open-model ecosystem in China, which provides users in the West access to highly capable systems for free. This openness reflects a strategic choice by Beijing and the firms in its orbit. By driving the easy adoption of its systems — from startups in Malaysia to even a US federal agency — China exports technological influence abroad. Beijing also controls only around 15% of the world’s high-performance AI computing power, compared to about 75% in the United States. This means it has far less capacity to serve its models to customers at scale and capture the revenue that doing so would provide. Related Article Block Placeholder Article ID: 1267800 Should we replace politicians with AI? Would anyone notice? Bernard Keane 33 Yet as China’s systems become more capable — and their outputs more valuable — Beijing and its leading tech companies will have stronger incentives to limit how their best systems circulate abroad. The ready availability of Chinese open models may prove a temporary feature of the current bout of AI competition and not a mainstay on which Australia can reliably depend. The task for Australia is to manage the dependencies that AI will bring. The goal should not be to eliminate them entirely: developing truly sovereign frontier systems at home will be exceedingly expensive and is unlikely to produce any durable strategic advantage. The better strategy is to make our dependence harder to exploit by ensuring Australia has enough alternatives and bargaining power to avoid being exploited when the geopolitical winds change. The committee’s report must grapple with the geopolitical structures that determine who gets access to the leading AI systems — and who is most exposed to their risks. Australia is unlikely to become technologically self-sufficient and so must accept a serious degree of dependence. How this dependence is managed will determine whether Australia can reliably access the technologies shaping economic and strategic power.