Washington flicked a switch and Australian businesses lost access to some of the world's most powerful AI models. The fact it was reversed doesn't make the lesson any less confronting.
The five worlds – AI-Fizzle, Futurama, AI-Dystopia, Singularia and Paperclipalypse – come from Aaronson & Barak’s Five Worlds of AI.
This week, the US government briefly blocked Australian organisations from using Anthropic's most advanced AI tools – then lifted the ban. It's a reminder that the AI most businesses rely on is controlled by a small number of powerful companies and governments far away. That's closer to the scary Paperclipalypse future – where a handful of players hold all the power – than the hopeful Futurama future where AI is a tool everyone can use fairly. Meanwhile, businesses around the world are getting hit with AI bills much bigger than expected, and some of the world's top economists are warning the boom might not last. The picture is complicated – and the gap between who benefits and who pays is widening.
On Tuesday, Anthropic's Fable 5 came back online for Australian users. Most people treated it as good news. I think the more honest read is: we just got a live demonstration that the US government can switch off Australia's AI whenever it wants to – and then decide to switch it back on, on its own terms.
That single story, covered across 15 separate sources and easily the most widely followed event of the week, crystallised everything the numbers have been pointing to for months.
The weighted global score for this week puts AI-Dystopia at 31% – a sharp rebound from last week's relative calm at 24%, and back to levels we last saw in Scan 11 (35%) and Scan 10 (31%). The Distribution Gap – the gap between AI-Dystopia and Futurama – jumped from +1 last week to +8 this week. That's meaningful. It means the conversation about AI is, net, more about value concentrating in the hands of a few than about gains spreading broadly.
Futurama held steady at 23%, unchanged from last week. AI-Fizzle is 20% – near its series high of 22% from last scan. Singularia sits at 14%, consistent with its range. Paperclipalypse dropped to 11%, its lowest reading in the series.
This isn't noise. The Dystopia rebound after last week's dip is a continuation of the broader trend – Dystopia has been above 28% in ten of thirteen scans. Futurama has never breached 27% in the weighted global figure. The Distribution Gap has been positive (Dystopia leading) in eleven of thirteen weeks.
The structural story hasn't changed: AI's value is reading as concentrating upward, not spreading out.
Australia and New Zealand produced 66 signals this week – the region's busiest week – and the mix tells a sharp story. AI-Dystopia and Futurama are virtually tied at 26% and 27% respectively. The regional Distribution Gap is -2 this week (Futurama narrowly leads), consistent with its rolling four-week figure of -3.
So why does it feel more dystopian than the numbers suggest? Because the stories carrying the most weight are about power, not tools.
The Anthropic export control saga dominated ANZ coverage. SmartCompany's "Neural Notes" column put it plainly: the US keeps proving it can switch off Australia's AI. Information Age noted that Australia regained access to Fable but not Mythos – a partial restoration that actually underscores the dependency. Former PM Malcolm Turnbull called on Australia to stand up to Washington on AI. The Startup Daily published a genuinely good-faith piece arguing sovereign AI doesn't make sense for Australia economically – which is true and makes the dependency even harder to resolve.
On the AI-Fizzle side, the bill-shock cluster was significant. The AFR's corporate revolt story, backed by the SMH's tokenmaxxing piece, iTnews, and Cyber Daily, showed Australian and global businesses getting hit with AI spend they didn't budget for. Gartner's CHRO warning about hidden AI workforce costs added institutional weight to that concern.
The genuinely good news from ANZ: Bendigo Bank's agentic SOC ambition, Woolworths remaking its loyalty assistant, KFC operator Collins Foods trialling AI for margin improvement, Heidi shaking up NZ healthcare, and ABS data showing surging AI adoption among Australian SMEs. These are real, grounded adoption stories – not vendor promises.
North America ran 111 signals and produced the week's starkest concentration story. AI-Dystopia sits at 34%, Singularia at 26% – the region's highest Singularia reading in the series. Futurama is just 14%.
The Distribution Gap for North America is +20 this week, down slightly from its rolling +24 but still by far the highest of any region. If you want to understand where the global Dystopia reading comes from, look at North America.
The Anthropic export control story drove the Dystopia reading. Amazon's $1B Forward Deployed Engineers programme, Together AI's IPO preparations, the Crusoe $3B raise, SAP cutting hiring to fund AI, Oracle's widening layoffs, and a steady drumbeat of hyperscaler infrastructure stories all reinforced the concentration theme. Palantir's AI sovereignty manifesto – a vendor voice, noted – landed in this context and read less like marketing and more like a symptom.
The Fizzle signals from North America were serious. The Wall Street Journal ran top economists sounding the alarm. UBS reported the majority of enterprise companies it talks to are throttling AI spend. Ford's grey-beard rehire story cut through – a concrete, specific example of AI failing to deliver in industrial quality control. MIT Technology Review's LLM groupthink work added a technical layer to the doubt.
The Singularia spike is real but weighted toward practitioner and vendor voices: Cisco deploying agents to 90,000 employees, Cursor's software factory model, self-improving agent loops from Anthropic's research community. Treat this as directionally interesting, not yet an adoption mandate.
Europe is the closest thing to equilibrium in the scan – 32% Futurama, 32% AI-Dystopia, a Distribution Gap of zero this week against a rolling -4. The HCLTech $1.14B deal, Microsoft Excel AI tools for finance, skills academies, and the DeepMind alumni quant AI story anchor the Futurama reading. Schneider Electric's $3.1B Cognite acquisition, Oracle Romania layoffs, PyTorch sovereignty framing, and the AI Kings Cross clustering story push Dystopia. Europe is watching both sides and governing actively.
Singapore and Hong Kong contributed 62 signals. AI-Dystopia leads at 37%, rolling at 42%. The Kling AI $3B raise (Kuaishou/Tencent), ByteDance's sovereign CPU ambitions, China's largest model trained on domestic chips, Nvidia's robotics hiring expansion in Beijing and Shanghai, and the Nvidia chip fraud case in Singapore all tell the same story: compute and model capability are concentrating fast, with geopolitics doing much of the work. Singapore's data centre regulation bill and OCBC's avatar banking rollout offer genuine governance and adoption counterweights.
Economy is the dominant force behind both AI-Fizzle and AI-Dystopia this week. Bill shock, budget throttling, hyperscaler capex concentration, data centre infrastructure economics – these are all economy-driven. Geopolitics is the second engine, almost entirely in Dystopia: export controls, sovereign AI debates, China-US compute competition, ByteDance's chip strategy. Technology drives Singularia and much of the infrastructure Dystopia. Trust underpins Paperclipalypse – shadow AI governance gaps, ungoverned agents, AI-supercharged cyber threats. Talent splits between Futurama (skills programmes, reskilling) and Fizzle (skill decay, shallow adoption, workforce readiness declining despite adoption rising). Regulation is present but reactive – responding to concentration rather than shaping it.
Almost nothing this week came from professional services firms actually implementing AI at scale. Healthcare adoption – outside Heidi and Novant – was thin. SME voices were nearly absent despite ABS data showing they're adopting fast. And there's a striking silence on what organisations are actually measuring when they call AI a success. "Adoption" is everywhere. Verified, audited productivity gains at the business unit level? Scarce.
Here's the thing. If your AI strategy runs through a single US-based vendor and you haven't mapped your exposure, the Anthropic episode just ran your risk scenario for free. The question isn't whether to use frontier AI. It's whether you've thought about what happens when the access changes – pricing, availability, policy – on someone else's timetable.
The bill-shock cluster is the other Monday morning story. If you're not actively governing AI spend at the token level, you're not governing it at all. The organisations throttling spend aren't retreating from AI. They're getting smarter about where the value actually is.
Find that. Own it. Don't let someone else's infrastructure decisions be the thing that decides it for you.
The articles most responsible for bending the five worlds this week.
The US government briefly cut Australian businesses off from Anthropic's most powerful AI tools, then reversed the decision. This column made the point that matters most: the reversal doesn't fix the problem. A handful of American companies and policymakers control the AI tools Australian organisations depend on, and that's a risk most leaders haven't properly planned for.
Companies that jumped into AI are now getting hit with bills far bigger than expected, and some are pushing back hard. This story – backed up by similar reports in the SMH, iTnews and Cyber Daily – showed that the cost-versus-value maths isn't working for many organisations yet. When the CFO gets involved, AI projects get scrutinised properly for the first time.
It's one thing for sceptical tech commentators to question the AI boom. It's another when the world's leading economists start publicly voicing concern that the promised economic gains may not materialise. This story gave serious institutional credibility to the idea that the AI investment wave could be outrunning the actual value it delivers.
Nine thousand jobs gone, with AI cited as the driver. This is the real-world version of the abstract debate about who benefits from AI productivity gains – and this week the answer was clearly the company's shareholders, not its workforce. Stories like this are what push the gap between 'AI helps everyone' and 'AI helps the top' wider.
The people most likely to bypass an organisation's AI rules aren't junior staff – they're the C-suite. This report found senior leaders are using AI tools that sit outside approved systems, creating data and security risks that nobody is auditing. When the people setting the governance rules are also the ones ignoring them, the governance isn't really governance.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free