There was a revealing moment last month when Communications Minister Anika Wells appeared on the ABC’s Insiders to discuss the government’s latest duty of care provisions for social media. Asked about whether the government wanted AI development to slow down, she let slip an astonishing aside. “AGI is here, isn’t it?”, Wells said, matter-of-fact, referring to “artificial general intelligence”. The statement was not widely reported. An ABC write-up even changed Wells’ quote to remove the “G”, changing her remark to simply, “AI is here, isn’t it?” AGI — a still-hypothetical evolution of AI that could learn, reason and adapt across a broad range of tasks with limited human oversight — is the nub of the idea behind Donald Trump’s favoured term, “super intelligence”. Its existence is also fierce topic of debate among cognitive scientists, psychologists and technology researchers, because AGI’s emergence would herald a profound ethical, social and economic crisis for humanity. Perhaps Wells was echoing Nvidia boss Jensen Huang, who last month claimed “AGI has arrived”. But regardless, her quip tells us something important about an ambient atmosphere of AI adoption pervading the Canberra policy discourse, one heavily influenced by the trillion-dollar hype machine of Silicon Valley. 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. Because even if “AGI” isn’t “here”, the government is sure acting like it is. ‘Drinking the Kool-Aid’ Anthony Albanese likely wasn’t anticipating AI to be one of the key policy challenges of his government. But the rapid intensification of the industry in recent years has made that inevitable. Once a decidedly niche topic parcelled out to the boffins and mid-level bureaucrats, AI policy has moved to the centre of political discussion. Internal government inquiries culminated in the release of the National AI Plan last December. Its framing is best summed up by the title of its first chapter, “Capturing the opportunities”, which breezily suggests “Australia’s success in capturing the opportunities of AI depends on equipping our people to get the best out of AI in their jobs, communities and their personal lives”. The document canvasses a concerted effort to “attract investment” and only gets around to emerging risks from new technologies towards the end. Anthropic co-founder and CEO Dario Amodei (Image: Reuters/Carlos Barria) Optimism about AI continued in early 2026, boosted by a visit to Canberra from Anthropic’s Dario Amodei. Presenting himself as the smiling face of ethical AI research, Amodei was at the time basking in the glow of his company’s successful Claude model. With an entourage of flunkies and nerds, the CEO secured meetings with Albanese, Treasurer Jim Chalmers and Assistant Minister for Science, Technology and the Digital Economy Andrew Charlton, while also hosting a parliamentary forum that became a hot ticket for Labor backbenchers and public service top brass. A senior operative in a Labor minister’s office described Amodei’s reception to Crikey as “drinking the Kool-Aid”. The CEO’s rock-star treatment partly stems from the rapid uptake of chatbots and large language models (LLMs) by Canberra elites themselves. There can be no denying the proliferation of these tools in the senior public service and among many policy, industry and academic elites. ChatGPT’s record run from early 2023 is widely recognised as the fastest-adopted consumer application in history. This AI hype, which has helped cloak the naked avarice of Silicon Valley, has also infected large parts of Australia’s managerial and professional elite. The knowledgeable chattiness of Claude or ChatGPT seems particularly attractive to a certain class of senior executive, university vice-chancellor and departmental secretary. Resembling an unusually bright and efficient research assistant, AI chatbots are ever ready with a helpful summary, an evidence base, or a pre-formatted table with all the information laid out just so. Such are the disordering effects of this mania that even senior union leaders like the ACTU’s Joseph Mitchell appear to have fallen for the snare. Mitchell has co-authored a statement with a number of university AI departments calling for “a government-coordinated effort bringing together unions, universities and industry experts, the superannuation sector, and the Australian tech industry to … build our own AI models.” Apparently, the turkeys want to start modelling for Christmas.  Quid pro quo For the econocrats advising Jim Chalmers in the Treasury, AI has emerged as the long-hoped-for productivity saviour that will spur a generational boost in economic outcomes. Treasury’s recently released intergenerational report begins with a chapter devoted to “the AI revolution”, with glowing descriptions of “a general-purpose technology like steam power, electricity and the internet” that is “expected to have a profound effect on our economy and society”. AI is presented as a technological panacea, with “the potential to make essential services more accessible, affordable and responsive to individual needs”. While admitting some jobs might be lost, the report is remarkably optimistic about the longer-term effects, arguing that “as AI becomes widely adopted, it is expected to drive increases in productivity and incomes, supporting lower prices and higher consumer demand and employment”. This is the common sense underlying the breezy techno-optimism of the Productivity Commission and Treasury. It’s aptly summarised by Stephen King and Julie Abramson, who, when they were PC commissioners, wrote in The Australian last year that “technological shifts — from computers to the internet — are partly why the average Australian today has living standards three times higher than the average Australian in the 1960s”. In this worldview, pesky laws and regulations like copyright are impediments to the full deployment of AI throughout the economy. Treasurer Jim Chalmers speaks to Prime Minister Anthony Albanese during question time (Image: AAP/Lukas Coch) Artists and the cultural sector are fighting tooth and nail against such exemptions, which would allow AI companies to train models on creative works without compensation. Anthropic and OpenAI are seeking the carve-outs for legal protection. Having enthusiastically pirated material to train their models, they’re now facing the fallout: Anthropic, for example, recently forked out US$1.5 billion in a settlement to authors and book publishers. In the wake of Amodei’s visit in late March, creatives descended on Canberra in a last-ditch lobbying effort to fight back. The government issued suave assurances that it wouldn’t change copyright law. That hasn’t stopped intense lobbying pressure from Anthropic and OpenAI as they negotiate behind the scenes. Indeed, the temptation of a quid pro quo — copyright carve-outs in return for data centres and AI training — will be strong in Canberra, lured by the prospect of billions in investment and the “sovereign capability” of a home-grown frontier model. “Australia simply cannot afford to miss this AI wave”, enthused Peter Hartcher in a column last month, explaining how Australia has the opportunity to secure access to frontier models by allowing favourable regulatory treatment for Anthropic and OpenAI. ‘Do a deal’ This idea, that the Albanese government should “do a deal” with frontier AI firms on copyright, has been taken up with gusto by Liberal Senator James Paterson in a series of media interviews. For what, exactly, and for what in return, has not been spelled out. “Anthropic and OpenAI have the most advanced models in the world, and what we should be seeking to ensure is that we have sovereign access to those models,” Paterson told News 24’s Andrew Clennell on September 27, claiming that this could be done by encouraging the AI firms to spend more on training in Australia, “which will effectively guarantee our access to these models”. Reasonable questions might be posed about this putative deal, not least of them whether Paterson and his colleagues in parliament really understand what a “frontier model” is. How would this deal be done? What does “access” actually mean? Anthropic’s much-ballyhooed memorandum of understanding with the Australian government, signed after Amodei’s triumphant swing through Canberra this year, is a vague nothingburger. It makes some airy promises about “technical exchanges and collaborations with safety and security institutions” and mentions “Anthropic intends to take steps towards ensuring benefits to local industry and researchers through access to its infrastructure”. The memorandum is not a binding deal, stipulating “this MOU is intended as a statement of intent and is not intended to have legal effect.” Like so much of the AI policy discourse, this MOU reveals more about the imaginative capture of Canberra’s elites than it does about the supposedly historic opportunities or existential threats of emerging technologies. Little scrutiny appears to have been applied to the sales pitch. It doesn’t take a lot of imagination to understand why AI firms desperately chasing revenue to justify their sky-high valuations would be more than willing to give Canberra bureaucrats a test drive on their latest products. Nor does it take the processing power of Claude to realise the windfall that a copyright carve-out would bestow. The national security argument is only slightly more credible. Perhaps hyper-capable hacking agents are running on special US government servers somewhere. Or maybe the Australian Signals Directorate needs a special agentic goalkeeper that it can’t build itself. But it’s not clear that “access” to the bleeding edge of model development is a prerequisite for Australia’s national cybersecurity, even assuming Australian spy agencies have the capability and operational skill to effectively use such models. Similarly, arguments that data centres physically located in Australia can somehow be controlled by the sovereign local jurisdiction seem to display a touching faith in the probity of the firms operating them. On borrowed time and money At a more prosaic level, there must be open questions about whether any sovereign government can trust the AI billionaires. Voters are being asked to believe that a Commonwealth that can’t manage to tax foreign-owned gas companies will somehow transform into a ruthless negotiator for frontier modelling access. The Albanese government, egged on by Angus Taylor and Paterson, appears to be suggesting it can strike favourable hosting-for-access deals with the same frontier AI company that has just admitted to hacking into Australian government servers. As they used to say in the old days, it doesn’t compute. It can’t be a surprise that ordinary citizens are worried about a large and powerful industry implementing a massive project of surveillance, automation, pollution and extraction. The intensifying opposition to rampant data centre construction, both in the US and increasingly in Australia, is inextricably linked to growing anxiety about the future risks and impacts of technological change. The industry’s response has been to spend up on extra lobbying efforts, while most of the states and territories are still panhandling for every new data centre they can attract. The accelerating opportunity costs reach their full realisation in the hyperscaled data centre buildout, which has catalysed what can only be described as a fully blown speculative bubble. In a time when inequality is soaring and billions lack basic human security or access to clean drinking water, a vast mission investing in non-human succour and support has been inaugurated that draws on much of the world’s available liquid capital. According to recent data compiled by Stijn Van Nieuwerburgh for the Brookings Institution, the data centre build-out rivals the investment poured into some of the largest booms in history, larger than the canals and cotton mills of industrial revolution-era Lancashire, the American railway boom of the nineteenth century, or the post-war American interstate highways financed by Eisenhower. Much of this money has been borrowed. But the medium-term revenues needed to repay these loans appear uncertain, to say the least. On screen, Anthropic CEO Dario Amodei (L) and Hugging Face CEO Clement Delangue (R), with OpenAI CEO Sam Altman, speaking at the United Nations Security Council meeting on AI, September 23 (Image: CNP/Kyle Mazza) Perhaps investors really do believe that superintelligent machine gods can replace the white-collar workforce, vacuuming up swaths of the global economy’s daily spending in the process. But the current bubble also has all the hallmarks of past manias, in which a financial sector in the grip of a speculative frenzy piles into ventures with only the flimsiest prospectuses. The numbers can make the eyes glaze over, but some tell a story of their own. In August, Alphabet sold US$5.5 billion of corporate bonds in Australia, which followed huge bond programs in Canada. Why is a company like Alphabet — until a couple of years ago, a corporation swimming in free cash flow — suddenly tapping smaller bond markets, borrowing billions of dollars in foreign currencies at coupons above 6%? The answer is that Alphabet, like much of the rest of Silicon Valley, is investing stupefying amounts of money into data centres: more than US$200 billion this financial year alone. Alphabet is rich, with a long history of successful technology builds. But a whole class of lower-tier data firms has sprung up, with business models nothing more complicated than promising to build data centres with money they have yet to secure. These are the so-called “neoclouds”, atmospheric corporate formations that have borrowed eye-watering amounts to build or lease data centre capacity, which they then sell on to tech firms and other corporates looking to get access to advanced processing units. Australian neocloud Sharon AI is an eyebrow-raising local example. Helmed by James Manning, Sharon is listed on the NASDAQ and claims to control 212 MW of data centre capacity, most of it apparently leased from NextDC in Australia. The firm has borrowed more than $1 billion in convertible notes and recently inked an oversubscribed $1.6 billion private funding round. It appears to be using this debt to lease Nvidia AI chips. Despite its lofty ambitions, Sharon AI loses money. According to its second-quarter financial results, the firm booked just US$1.9 million (with an “m”) in revenue in the three months to June 2026, for an operating loss of US$12 million. It doesn’t appear to actually own any data centres, listing just US$26 million on its balance sheet for “property and equipment”. The majority of Sharon AI’s “contracted revenue” is with a mystery US$950 million customer, whose name Sharon won’t reveal. But Sharon is a picture of conservatism compared to its highest-profile local competitor, Firmus. Already notorious for its breakneck development of Tasmanian data centres, Firmus is careening towards an IPO this month that some analysts believe could be one of the largest in the history of the Australian stock exchange. Like Sharon and the American neoclouds, Firmus’ business strategy is a mix of vertical growth forecasts and circular financing. Its key supplier is Nvidia, which is loaning it money to buy its chips. Its key customer is OpenAI. Despite this, Firmus has very few actual chips, little currently existing data centre capacity, and no publicly disclosed audited financials. Oh, and one of its founders, Oliver Curtis, is a convicted criminal who spent time in prison for conspiracy to commit insider trading. Unsurprisingly, opposition is growing in Tasmania, where Firmus plans to suck up one-seventh of the state’s existing electricity supply.  You don’t have to be Charles Kindleberger or Hyman Minsky to see some looming risks. By Thursday, lurid rumours were circulating in the business press that Firmus’ IPO was in trouble, with some speculating that the firm’s brokers would abandon the initial offering altogether. ‘Jobs and growth’ Ask an ordinary Australian what should be built with a historic infrastructure program, and it’s a fair bet that “data centres” would be well down the list. Most of the nation’s infrastructure needs are social, not digital. Australia needs houses, aged care beds, affordable health care, schools for our children, potholes fixed and trains for our sprawling outer suburbs. We are only halfway through a historic transition from an expensive and polluting coal-fired electricity grid. All of these requirements are competing with the skills, materials, and especially the finance currently being poured into server racks for Nvidia graphics processor units. Perhaps the most unsettling aspect of the AI bandwagon is the degree to which significant investment decisions about the future of the economy are being made by foreign billionaires, with little consideration of the needs or desires of ordinary Australians. None of this seems to be worrying Australian politicians, who are falling over themselves to welcome new data centre investment, even as they claim to promise strict new safeguards on new data centre development. Tasmanian Premier Jeremy Rockliff has been an enthusiastic spruiker for Firmus, which is sponsoring the new Tasmanian Devils AFL football franchise, while in New South Wales, Premier Chris Minns has gone so far as to claim that the state would be in recession if it weren’t for data centre spending. Prime Minister Anthony Albanese and NSW Premier Chris Minns at a press conference following a national cabinet meeting, August 26 (Image: AAP/Dean Lewins) The endearment of politicians for “jobs and growth” is one reason for the vast gulf that has opened on AI between elite and popular opinion. While the technocrats dream of limitless opportunities at the modelling frontier, ordinary folk can’t help taking note of the rhetoric of job destruction and human extinction spouted by the Altmans, Amodeis and their mid-level San Francisco lackeys. Companies laying off staff are now routinely claiming AI is responsible, including Atlassian in Australia, while the media has enthusiastically covered both recent “rogue agent” security attacks and the wilder claims of the effective altruist doomerists. Mild-mannered suburbanites are banding together to fight the data centres springing up in their backyards, while none other than the Reserve Bank is blaming the data centre spending for hiking interest rates. Sociological, psychological and humanitarian theories about technology help us understand the strange times we are living through. Powerful new technologies are profoundly political inventions, even as their broader impacts are socially and economically disordering. Australia’s political class has consistently struggled to come to terms with emerging technologies and has typically chosen to respond to the challenge by outsourcing its policy ideas to tech giants and consultancies, and AI is only the latest case. What’s at the heart of Albanese’s project By mid-2026, the debate about AI had reached a stage where Anthony Albanese clearly felt the need to make a public statement, if only to kick the can down the road until some actual regulations could be drawn up. In July, the prime minister gave a speech staking out his position. Entitled “AI in Australia’s interests”, Albanese attempted to balance the opportunities and threats of AI while stressing a frame of national interest. “Our power, our agency, our choice lies in embracing change and shaping it,” he argued. “Not just adopting or accommodating AI: designing it, making it, building the capability right here.” In the speech, Albanese committed to safeguarding copyright and putting in place “the strongest possible protection for Australian artists and Australian media”. Weeks later, ACT independent Senator David Pocock tabled confidential consultation documents proposing an “opt-out” copyright exemption for model training, which many in the creative industries suspect was secretly negotiated with OpenAI and Anthropic. An “opt-out” clause would allow technology firms to automatically use copyrighted material for training, unless artists and copyright holders explicitly object, creating a loophole that would allow AI firms to legally use scraped content for their training. This has kicked off a round of anguished op-eds from artists and authors like Richard Flanagan, but the hand-wringing may not be enough to stop the AI lobbying juggernaut. In her report for The Saturday Paper this week, Karen Barlow quotes Labor backbencher Ed Husic (a former Industry minister himself) describing a “shock and awe lobbying approach by US tech, designed to swamp government decision-making”. Kate Pounder, a former boss of the Tech Council of Australia and one of Andrew Charlton’s former colleagues in the lobbying firm AlphaBeta, has just been appointed by the Malinauskas government as one of the commissioners for its state-based royal commission on AI. Pounder has also worked for OpenAI. It’s not looking good for the artists. The inside games shed further light on why Albanese and his government are still talking to Sam Altman, even after an OpenAI agent gained unauthorised access to a Medicare portal earlier this year. OpenAI has been less than forthcoming on the incident: Altman met with Deputy Prime Minister Richard Marles in early September, where he somehow forgot to mention the attack. Senior executive Jason Kwon, who’s been sent out for a round of apologies, claims Altman didn’t know about the hack when they met. Regardless, the incident doesn’t seem to have fundamentally altered Labor’s enthusiasm for frontier model access.  OpenAI chief strategy officer Jason Kwon during the Joint Select Committee on Artificial Intelligence, Sydney, October 6 (Image: AAP/Sarah Wilson) The government’s indecision about whether to regulate AI mirrors a broader confusion at the heart of Albanese’s project. Since 2022, Labor has attempted to implement a new style of industry policy, financing a large investment fund for a “Future Made in Australia”, and bailing out this or that smelter or refinery. But there seems to be little in the way of overarching economic strategy, and most of the fundamental settings are the same neoliberalism of the past 40 years. In this context, the government seems to be treating AI like a digital version of the mining boom — a comparison drawn by Charlton in a speech to ANU in August. “We need data centres, but only the good ones,” he claimed last month. AI policy shows how Labor’s broader social project under Albanese has slowly hollowed out. The party appears too estranged from its eroding electoral base to respond to the clear desire from many working Australians for a pause in the AI hype-doom loop. A government in the sway of tech lobbyists, entwined in the US national security apparatus, moving rapidly to implement new algorithms to determine aged care access and making draconian cuts to disability care is not likely to be one that will regulate AI in the best interests of ordinary Australians.