A US government order cut Australian organisations off from Anthropic's most advanced AI overnight. The question every CEO needs to answer: how many of your critical systems run on someone else's permission?
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 told Anthropic – one of the world's most powerful AI companies – to stop letting people outside America use its best AI models, and Australia lost access immediately. Think of it like the US owning the only power station in town and deciding to turn off the electricity to foreign customers. That's the scary AI-Dystopia direction: a world where a handful of powerful countries and companies control who gets to use the best AI, and everyone else has to ask nicely. The hopeful Futurama direction – where AI is a tool that helps everyone and spreads the benefits around – was still visible this week in stories like Bunnings selling through Google AI and NAB improving customer service with smart data tools. But right now, the scary direction is winning.
Somewhere in Washington last week, a government official signed an order and Australians lost access to Anthropic's most powerful AI models – Fable 5 and Mythos 5 – overnight. No warning. No transition plan. Just: gone.
That single event generated 36 separate stories in this scan. It's the most-covered story in the series so far, by a distance. And it didn't just dominate the news cycle – it shifted the numbers in ways that should make every CIO in Australia sit up straight.
This week's global Distribution Gap – the margin by which AI-Dystopia signals (value concentrating in a few powerful hands) outrun Futurama signals (value spreading broadly) – hit 11 points. That matches the series high set in scan 10, the week before. Two consecutive weeks at 11 is not noise. It's a trend.
Globally, AI-Dystopia claimed 35% of all signals this week – a series high. Futurama held at 24%. AI-Fizzle ticked up to 20%, its joint-highest mark. Paperclipalypse fell to 11% – tied for the series low – which tells you something: the debate this week wasn't about AI running out of control. It was about who controls the controls.
The five worlds, for readers new to the framework: AI-Fizzle is the scenario where AI fails to deliver on its promises. Futurama is where it works and everyone shares the upside. AI-Dystopia is where it works but the gains concentrate at the top. Singularia is where AI capability accelerates beyond what humans can fully manage. Paperclipalypse is where AI causes harm that nobody can contain. The Distribution Gap – Dystopia minus Futurama – is the key instrument. Positive means value is concentrating. Negative means it's spreading.
At 11, it's pointing the wrong direction.
Here's where it gets personal. Australia and New Zealand carry 40% of the weighting in this scan because that's who we're writing for – and the ANZ picture is complex in a way that demands attention.
This week's ANZ Distribution Gap is +6 (Dystopia 35%, Futurama 30%). That looks manageable. But the rolling 4-week figure tells a very different story: −8. That means over the past month, ANZ signals have actually leaned Futurama. This week's shift is sharp and sudden – driven almost entirely by the Anthropic shutdown and its Australian coverage. The AFR's "fears of an AI Strait of Hormuz" framing, The Conversation's "the US government can shut off access to AI at will", Capital Brief's "Australia has stronger AI bargaining chips than we realise" – all of it landed in the same week, and all of it is Dystopia.
The good news, and there is some: the Futurama stories are real and local. Bunnings selling through Google AI Mode. NAB using Databricks to improve dispute resolution. MinterEllison creating a Chief AI Officer role. NZ's Heartlab signing a major cardiac AI deal reaching 500,000 patients. A $65M Series A for Melbourne health-AI startup Everlab. These aren't press releases – they're genuine adoption with humans in charge and value spreading.
But they got drowned out this week. And the Fizzle signal is worth watching: Amazon CTO Werner Vogels warning too many organisations adopt AI out of fear, leadership falling behind AI momentum in Australian workplaces, the confidence gap holding back NZ small business. The gap between intent and execution is real and persistent in this market.
North America ran a Distribution Gap of 22 this week – nearly double the global figure. That's been structurally high all series (rolling: 18), and this week explains why: the Anthropic shutdown, SpaceX acquiring Cursor for $60 billion, Nvidia raising $20 billion in bonds, Noam Shazeer moving from Google to OpenAI, DeepSeek closing a $7.4 billion raise at a $50 billion-plus valuation, the White House's AI power centre operating with enormous informal influence. The AI industry talking about itself dominated the NA lens – and the NA industry stories this week were almost uniformly Dystopia.
Separating industry voice from adoption evidence matters here. Jeff Bezos predicting labour shortages rather than mass layoffs is a Futurama signal – but it's still one CEO's prediction, not deployment data. The Salesforce acquisition of agentic AI firm Fin for $3.6 billion and ChatGPT's Visa integration are Singularia – they assume autonomous capability arriving. When you strip vendor voices and industry commentary and look only at organisations actually using AI, the North American adoption picture is more mixed: some real wins in financial services and retail, real fizzle in cost overruns, and real concern in the workforce.
Europe's Distribution Gap flipped negative this week: −3. Futurama matched Dystopia at 23% each, with Singularia surging to 20% and Fizzle at 23%. That's a very different reading from the rest of the world.
Some of this is structural: European coverage skews toward policy, governance and startup investment – all of which tend to land in Futurama or Singularia. Ireland passing the EU AI Act supervision bill, Estonia proposing government IDs for AI agents, French startup Conduct raising €60M from Palantir alumni – these are governance-engaged stories. The rolling 4-week Distribution Gap for Europe is −2, meaning the pattern holds. Europe is genuinely closer to a Futurama reading than anywhere else.
The Anthropic story hit Europe hard in framing – Sifted's "America has hit the AI kill switch, how should Europe respond?" was widely read – but Europe mostly responded with "this is why we need our own." Mistral's CEO pitching open-source sovereignty days after the ban is a telling data point.
APAC (Singapore, Hong Kong, China) carries only 5% weighting but ran a Distribution Gap of 29 – the highest in the scan. AI-Dystopia dominated at 48%. This is driven by China's AI arms race (DeepSeek, ByteDance, Alibaba, MLCC supply crunches, chip sovereignty debates), Singapore's two-tier AI salary market and the Anthropic shutdown's regional implications. The rolling figure (21) shows this isn't a one-week spike.
Geopolitics (45 signals) and economy (72 signals) are the two primary engines behind AI-Dystopia this week. The Anthropic shutdown is pure geopolitics – export controls applied to AI model access in a way that nobody's legal framework was designed for, as Computerworld's analysis of "what is an export?" laid bare. Economy is driving concentration in infrastructure: data centres, chips, GPU financing. These two forces fed each other all week.
Technology (79 signals) is the main fuel for both Futurama and Singularia – genuine capability arrivals like GLM-5.2 matching frontier performance, Alibaba's embodied robotics models, Salesforce's agent acquisitions. Trust (46 signals) split across Fizzle and Paperclipalypse – 16% of Americans believing AI will have a positive societal impact is a trust signal nobody in the industry should dismiss.
Talent (37 signals) is increasingly a Dystopia story: PwC's "seniorization" report showing entry-level roles morphing out of reach for young workers, tech workers spending nights upskilling just to stay employed, the two-track labour market hardening.
Notably absent from the discourse this week: the customer experience of enterprise AI at the mid-market level, any substantial measurement of ROI actually delivered (versus predicted), and almost any voice from frontline workers who aren't in tech. The narrative is dominated by CEOs, government officials, frontier lab founders and financial markets. The people doing the work in organisations are largely invisible.
Here's the plain reading for a CEO or CIO: the Anthropic shutdown proved that your AI stack is only as sovereign as the geopolitics allow. If you built a critical workflow on a US frontier model and assumed access was guaranteed, you now know you were wrong.
That doesn't mean panic and rebuild everything from scratch. It means doing something most organisations haven't done: mapping your AI dependencies the same way you'd map a critical vendor concentration risk. Which capabilities, which workflows, which customer-facing systems depend on a single model provider? What's your fallback? The week's most useful strategic insight came not from a frontier lab but from The AI Daily Brief: the enterprises that will survive model disruptions are the ones that have built institutional knowledge and model-portable processes – not just model subscriptions.
Bezos said labour shortages are coming. SAP's CEO said human coders have four years. The Anthropic shutdown showed the timeline can be zero days.
Ask the question before someone in Washington answers it for you.
The articles most responsible for bending the five worlds this week.
The US government ordered Anthropic to cut off access to its most powerful AI models for users outside America, and Australia was immediately affected. This was the week's defining story – it showed that a foreign government can effectively turn off AI tools that Australian organisations rely on, with no warning and no appeal process.
The AFR's "Strait of Hormuz" framing – comparing the AI access cut-off to a strategic chokepoint that could throttle global trade – captured exactly why Australian business leaders were alarmed. If the US can block AI access the way a navy blocks a shipping lane, the rules of the digital economy just changed.
SAP's CEO publicly said AI will replace the company's human programmers within four years. When the head of a company that builds the business software running thousands of organisations says this out loud, it's not a prediction – it's a signal about where enterprise software investment and workforce planning are headed.
Bunnings – the retailer almost every Australian knows – is now selling products through Google's AI shopping mode. It's a real, local example of AI changing how businesses reach customers, and it's the kind of adoption story that matters more than any executive prediction about the future.
Research this week confirmed that AI is splitting the workforce in two: people with the right skills pulling ahead, and everyone else falling behind. It's not a future scenario – organisations are already seeing it in hiring decisions and salary data, and it has direct implications for how CHROs and people leaders plan right now.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free