Abigail Bradshaw, head of the Australian Signals Directorate, told a parliamentary inquiry on AI on Friday that training artificial intelligence models on home soil is critical to safeguard national security. That’s the state saying the machines that now matter to defence and financial systems should be kept closer to the apparatus that watches, sorts, and controls them.
Bradshaw said hosting the training of frontier models could give intelligence agencies early insights into risks and greater control over access to AI models. The language is tidy. The power move is not. If intelligence agencies get early insights and tighter control, ordinary people get another layer of decisions made above their heads, in rooms they’ll never enter.
Who Gets the Control
The Australian Signals Directorate has nominated Anthropic and OpenAI as the preferred partners for the Australian government’s AI cyber defence efforts. The article said Anthropic and OpenAI were spruiked by ASD for Australian AI cyber defence. That’s the shape of it: public authority leaning on private firms to secure systems that already sit at the center of defence and finance, with the public left to trust the arrangement and absorb the consequences.
Bradshaw’s pitch rested on the idea that training models in Australia would help intelligence agencies see risks earlier and manage access more tightly. She framed that as national security. But the facts in the article show a familiar hierarchy at work: the people with the least say are the ones expected to live under the systems being built, while the agencies and their corporate partners decide how those systems operate.
Who Pays for the Deal
Anthropic has claimed its investment in Australia hinges on Labor’s decisions on copyright. The article said Labor is caught in a bind as Anthropic warns that the nation will miss out on a significant portion of its planned investment unless copyright laws are changed. So the bargain is laid out plainly. A corporation signals where it may or may not place its money, and the government is left weighing whether to alter the rules to keep the capital interested.
That’s not public control. It’s leverage. The company’s warning turns copyright into a bargaining chip, and Labor’s response sits inside the same old trap: adjust the law to suit the investor, or risk losing the promised cash. The people whose work and rights are bound up in those laws don’t appear to be the ones setting the terms.
What They Call Security
Bradshaw told the parliamentary inquiry that home-soil training is critical. She said intelligence agencies could gain early insights into risks and greater control over access to AI models that are now vital to defence and financial systems. Those are the systems the state says must be protected, and the article makes clear who gets to do the protecting: the Australian Signals Directorate, with Anthropic and OpenAI lined up as preferred partners.
The arrangement reads like corporate capture dressed up as preparedness. The state wants proximity to frontier models. The firms want favorable conditions for investment. And the public is told this is about safeguarding national security, as if the machinery of control becomes harmless just because it’s wrapped in that phrase.
The parliamentary inquiry on AI took place on Friday, the same day Bradshaw made her case. The timing matters because it shows the process in motion: the state consulting, the firms positioning, the rules around copyright becoming a pressure point, and the people affected by all of it watching from outside the room.
The article doesn’t describe any grassroots response, mutual aid effort, or horizontal organizing around the issue. What it does show is a closed circuit of power: intelligence agencies, major AI firms, and a government trying to keep investment flowing while managing the risks of the very systems it wants to deepen its grip on.