The US blocked global access to Anthropic's frontier models this week – and the Distribution Gap snapped back to its widest point in ten scans. The question every CEO needs to answer: who actually controls your AI strategy?
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 ordered Anthropic to shut down access to its most powerful AI models for users outside America, saying they were too dangerous to share. At the same time, OpenAI filed secret paperwork to list on the stock market at a valuation of over a trillion dollars. Together, these two events tell a big story: a tiny number of companies and governments are taking control of the most powerful AI tools in the world, and everyone else has to ask permission. That's the scary path – what researchers call Paperclipalypse territory, where AI causes harm no one can stop, and AI-Dystopia, where the benefits pile up for a few at the top while everyone else pays the costs. The hopeful path – Futurama, where AI stays a useful tool and the benefits spread widely – took a step backwards this week. The good news is that it's not too late to change direction, but only if leaders pay attention now.
I want to start with a number: 10.
That's the Distribution Gap this week – AI-Dystopia at 31%, Futurama at 21%, a positive gap of 10 points. It's the joint highest reading in this entire ten-scan series, matching the very first week back in April. After last week's rare negative reading of minus four – the only time Futurama actually led Dystopia in this series – the gap has swung 14 points in a single week. That's the biggest single-week reversal we've recorded.
This isn't noise. Two stories drove it, and they arrived within 48 hours of each other.
The US Commerce Department moved to restrict foreign access to Anthropic's Claude Fable 5 and the Mythos-class models it's built on, citing security risks. Anthropic's own safety warnings – the very guardrails the company spent years building its reputation on – appear to have triggered the intervention. A researcher demonstrated what looked like a defence-relevant exploit; the government acted. Anthropic pulled the models globally. Then the CEO called for government to block dangerous AI. Then Trump reignited the feud publicly. Twenty-six separate stories covered this chain of events. It is, unambiguously, the most covered AI story in this series.
At almost exactly the same moment, OpenAI filed confidentially for a US IPO, reportedly targeting a valuation north of a trillion dollars. Goldman Sachs and Morgan Stanley are reportedly battling for the mandate in what Fortune called a "$7 billion horse race." OpenAI is also pivoting ChatGPT toward a superapp before it hits public markets.
Read those two storylines together and you get this week's thesis: the most powerful AI in the world is being simultaneously locked down by governments and locked up by capital markets. The people who built it are calling for oversight. The people who fund it are racing to go public. And the rest of us – including most Australian and New Zealand businesses – are watching from outside the room.
The macro forces this week are unusually concentrated. Economy (89 signals) and technology (95 signals) remain the dominant drivers, but what's interesting is where they're pulling.
Technology is driving Singularia and Futurama – frontier model releases, agentic AI deployments, robotics fundraises. The capability story is genuinely accelerating. Anthropic's Mythos-class models represent a real leap; Ethan Mollick at Wharton wrote this week about what it actually feels like to work with them, and his framing is Singularia-adjacent.
Economy is driving AI-Dystopia – IPO filings, hyperscaler capex, price wars, infrastructure concentration. The financial architecture of AI is consolidating fast. Amazon borrowed $17.5 billion from banks for AI spending. Oracle's AI earnings blew past estimates. Applied Digital signed a $5.2 billion data centre lease with a single US hyperscaler.
Trust is the fuel behind Paperclipalypse this week – and it's burning hot. The Anthropic model pulldown, the KPMG hallucination scandal (the consulting giant retracted a report on AI after it was found to contain AI-generated errors), xAI firing an engineer who raised safety alarms about Grok, a Bank of England warning about AI deepfake scams, and a New Zealand FMA alert about AI-driven investment fraud totalling $265 million. Trust erosion is running across every sector, every region.
AI-Fizzle is being carried by talent and economy – the 6.4-hours-a-week babysitting bots finding from Computerworld, the "AI adoption spending spree is over" signal from CIO, only 14% of Australian firms having scaled AI per the Asana data picked up by iTWire, and the persistent pilot-to-production gap.
Regulation is doing real work this week – not just as a background force but as an active plot driver. The US export controls, the EU blocking Apple's Siri AI at launch, the EU antitrust order forcing Meta to open WhatsApp to rival chatbots, and the German court ruling that held Google liable for AI Overview results all landed in the same seven days. Courts and regulators moved faster this week than they have in months.
The ANZ Distribution Gap this week is minus one – Futurama (27%) just ahead of AI-Dystopia (26%). On the rolling four-week average, it's minus eight: Futurama has been consistently ahead of Dystopia here in a way it isn't anywhere else in the world. That's the structural difference of the ANZ conversation.
But the mood shifted this week. The Anthropic pulldown hit local tech media hard – iTnews, the AFR, the Sydney Morning Herald and Cyber Daily all covered it. The AFR ran a piece on critics warning AI safety is being overlooked in the sprint to go public. The SMH asked whether Trump and Xi alone can stop AI spiralling out of control.
Locally, there's a genuinely instructive split. On the Futurama side: Wesfarmers CEO Rob Scott on the Fear & Greed podcast describing practical AI rollouts across Bunnings and Kmart. Westpac embedding AI across core banking flows. Virgin Australia integrating ChatGPT into flight search. Clayton Utz becoming the first Australian firm to achieve ISO 42001 AI management certification. These are real deployments, human-directed, value-delivering.
On the Dystopia side: Canva slowing hiring amid AI pressure. ANZ IT workers facing pay stagnation and employment anxiety per ARN. Data centres overtaking breweries as Sydney's biggest water users. A one-gigawatt data centre approved in New Zealand amid national electricity shortages. The AFR's piece on how AI is getting suddenly more expensive as hyperscalers end developer subsidies.
Only 14% of Australian firms have scaled AI. That's the headline number. Everything else is context for why.
North America's Distribution Gap is 28 points this week – Dystopia at 39%, Futurama at just 11%. That's the dominant regional force pulling the global weighted average toward Dystopia. The rolling four-week gap is 17, so this week is a significant spike.
The Anthropic-OpenAI-SpaceX IPO trifecta, the hyperscaler infrastructure arms race, the data centre backlash spreading state by state, Anthropic's self-tax proposal acknowledging AI will destroy jobs at scale – these are all North American stories, and they all read Dystopia.
The one countervailing signal worth watching: Goldman Sachs forecasting 24x token consumption growth by 2030. That's not Futurama optimism – it's Singularia territory.
Europe's Distribution Gap is minus three – Futurama ahead of Dystopia on a weekly basis, which aligns with its rolling average of minus one. The EU AI Act guidelines, the Google liability ruling in Germany, the Meta WhatsApp antitrust order and the Siri block across EU markets are all real governance events, not just announcements. Europe's regulatory machinery is the most active it's been in this series.
The KPMG hallucination scandal – a global Paperclipalypse signal picked up strongly in European trade press – is the week's most important professional services story and a direct warning for any consulting or legal firm using AI to produce client-facing work.
APAC's rolling Distribution Gap is 17 – the highest sustained Dystopia-ahead reading of any region. DeepSeek close to sealing a $7 billion funding round. China's two-trillion-yuan national AI infrastructure plan. Moonshot AI eyeing a $30 billion valuation. Shopee cutting hundreds of developer jobs in its pivot to AI. Singapore private wealth getting selective on AI-era technology exposure.
The concentration story in APAC is about sovereign and corporate power pooling fast, and it's been consistent for four weeks.
Here's what I'd say to a CEO or CIO sitting down on Monday morning.
The Anthropic pulldown is not just a story about one company or one government. It's a demonstration of something your AI strategy has to reckon with: you are building on infrastructure that other people control, and those people can change the rules without warning. OpenAI's IPO paperwork is a related signal. When your most critical AI vendor goes public, its obligations shift – toward shareholders, toward markets, toward the government that just showed it can shut down its models overnight.
You need a second question running alongside "what AI can we build" – and that question is: "what happens to our operations if this vendor's access changes tomorrow?" Most Australian businesses haven't asked it seriously. The ones who have are the ones investing in governance, in vendor diversification, and in the human judgment layer that no model pulldown can take away.
The Distribution Gap is back at its series high. The value of AI is reading as concentrating, not spreading. That's a structural risk as much as a market signal.
The question isn't whether AI is powerful. This week proved it's powerful enough for governments to shut it down. The question is whether your organisation is in a position to make a choice – or whether you're just along for the ride.
The articles most responsible for bending the five worlds this week.
The US government ordered Anthropic to shut down access to its most powerful AI models for users outside America, saying they posed security risks. This was the single biggest story of the week – 26 publications covered it. It showed that even the most safety-conscious AI company in the world can have its technology pulled by a government with almost no warning, which is a wake-up call for any organisation that depends on these tools.
OpenAI filed secret paperwork to list on the US stock market, reportedly targeting a valuation of over a trillion dollars. This matters because going public means OpenAI's priorities will shift toward keeping shareholders happy, not just building safe or useful AI. For businesses that rely on OpenAI's tools, it's a reminder that you're renting from a landlord whose incentives are about to change.
KPMG had to pull a major report after it was discovered to be full of errors made by AI. This is embarrassing for one of the biggest consulting firms in the world, and it's a warning for any professional services firm using AI to produce client-facing work. AI can sound very confident even when it's completely wrong, and the people reviewing the work didn't catch it in time.
While global tech giants race to build AI infrastructure, Australia's own data centre boom is showing signs of stalling. This directly affects how quickly Australian businesses can access the computing power they need to run serious AI workloads locally. If the infrastructure isn't built here, the benefits – and the control – end up sitting overseas.
Only one in seven Australian businesses has actually scaled AI beyond small experiments – the rest are still stuck in pilot mode. This is the clearest local data point of the week on why Australia risks falling behind. It's not a technology problem; it's a people and process problem, and buying more AI tools won't fix it on its own.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free