This week's signals show AI's gains concentrating in fewer hands than ever, with regulators from Canberra to New York finally pushing back. The question is whether they're too late.
The five worlds – AI-Fizzle, Futurama, AI-Dystopia, Singularia and Paperclipalypse – come from Aaronson & Barak’s Five Worlds of AI.
This week, more stories than usual showed AI making rich companies and countries richer, while workers, smaller businesses and whole regions got left behind. That's the direction of a future called Paperclipalypse – where AI ends up serving the few, not the many, and ordinary people lose control over decisions that affect them. The better future, called Futurama, is where AI works like a really useful tool that everyone can access, with people still in charge. Right now, the numbers say we're closer to the scary version – but governments in Australia, the US and Europe are starting to write new rules to push things back toward fair.
Let me start with a number. Ten. That's the Distribution Gap this week – the difference between the share of signals pointing toward AI-Dystopia (33%) and those pointing toward Futurama (23%). It's the widest it's been since Scan 2 back in April, and it's back to the same level it hit in Scans 1 and 10. That's not noise. That's a pattern reasserting itself.
The Distribution Gap is the centrepiece of how I read this scan. A positive gap means the evidence says AI's value is concentrating – in platforms, in capital, in a handful of national champions. A negative gap means the opposite: the upside is spreading. At +10 this week, we're firmly in concentration territory. The question worth sitting with isn't whether that's happening – it clearly is – but whether the regulatory pushback that defined much of this week's story has any real chance of changing it.
The biggest story of the week in Australia wasn't about AI capability. It was about who gets to set the rules. Prime Minister Anthony Albanese's announcement of a new Office of AI and national framework – covered by 22 separate sources from the AFR to Crikey to The Conversation – dominated the local landscape. The policy pivot is real: mandatory grid connections for data centres, new AI standards, and the PM personally taking oversight of the government's AI agenda.
This is why Australia's Distribution Gap (week reading: +1) looks dramatically different from the global figure of +10. Locally, the regulatory response is being read as a genuine counterweight to concentration. But look at the rolling four-week figure and it tells a different story: -3, meaning the trend over the month has actually leaned toward AI's benefits spreading more broadly here than anywhere else in the scan.
The Albanese coverage split sharply in tone. The Financial Review called it a "major policy pivot." Crikey called it "AI FOMO" and questioned whether bureaucracy can actually deliver. The Conversation asked whether the missing piece was a sovereign capacity strategy – which it arguably is. What struck me most, reading across all 22 stories, was how little anyone disagreed on the diagnosis (AI infrastructure power is concentrating fast) and how much they disagreed on whether the cure would work.
The supporting ANZ signals matter too. A senior New Zealand policy adviser published a report populated with fake agricultural science generated by ChatGPT. Doctors in New Zealand are adopting AI faster than clinical governance can keep up. Eighty-one percent of Australian enterprises have already reported AI security incidents. A $2.26 billion deal to build NZ AI infrastructure landed with minimal scrutiny. And a mining giant told staff there'd be no new hires if AI could do the job.
Here's the thing: the Australian government is trying to set the rules before it loses the leverage. That framing, from a story in The Australian, might be the most important sentence written about AI policy in this country this year.
North America's Distribution Gap is 19 points this week – the most extreme reading across all four regions, consistent with a rolling figure of 20. If there's a place where the power-pooling story is most advanced, it's here.
The Apple-versus-OpenAI trade secrets lawsuit (14 stories, same story group) crystallised what concentration looks like at the frontier: two of the most powerful AI players fighting over IP, talent and the next hardware platform. The lawsuit threatens OpenAI's IPO timeline, which itself is a signal of just how much financial power is now riding on a handful of lab valuations.
Microsoft training its sales teams to actively undercut OpenAI and Anthropic (reporting in TechCrunch, TechRadar, The AI Daily Brief) tells you the hyperscaler consolidation play is now explicit, not implied. Satya Nadella warning of a "reverse information paradox" – where enterprises hand their proprietary knowledge to model vendors and can't get it back – is the kind of thing CIOs should be reading twice.
The IBM collapse (eight stories, same group) is the Fizzle counternarrative that deserves more attention than it's getting. A $69 billion one-day market cap loss because the AI boom is squeezing traditional software budgets isn't a story about IBM failing. It's a story about AI disrupting value before it creates equivalent value elsewhere. The SaaS apocalypse may be overrated, as ZDNET put it, but someone forgot to tell IBM's shareholders.
Europe's Fizzle reading hit 28% this week, the highest of any region, consistent with a rolling trend of 24%. The EU ordering Google to open Android to rival AI agents (same story group c35) is a regulatory story, but European scepticism about AI delivery runs deeper than regulators. The Barclays engineer warning that AI speed is a trap, the social atrophy argument from Esther Perel in Fortune, and Ireland's data centres consuming 23% of national power all point to a region asking harder questions about whether the promises are holding.
Europe's Distribution Gap (week: +16, rolling: -1) is its own paradox. This week looks sharp; the month looks balanced. The EU's regulatory moves are genuine checks on concentration, but whether they're fast enough is another question.
APAC's week reading shows Futurama (32%) ahead of AI-Dystopia (28%), with a negative Distribution Gap of -4. That's unusual, and it's largely driven by the volume of signals from China's World AI Conference in Shanghai. Xi Jinping's "AI for all" pledge, Moonshot AI's Kimi K3 release (largest open-weight model yet), and Chinese startups betting on phone-ready small models all read as spreading capability rather than concentrating it.
But read those signals carefully. Open-source and open-weight don't mean equally accessible. They mean accessible to whoever has the engineers, the data and the compute. And China's chip exports nearly doubling while Nvidia halves its Asia buyer list tells you the real geopolitics of compute are playing out underneath the rhetorical generosity.
Six macro forces shape this week's scan. Economy is the dominant engine behind both Fizzle (IBM's collapse, AI bills squeezing software budgets, hidden agent costs) and AI-Dystopia (hyperscaler capex concentration, DeepSeek's $74B valuation, Databricks at $188B). Technology fuels Futurama (open-weight models widening access, TCS scaling 8,900 deployment engineers, Bank of America's CFO reporting real productivity gains) and Singularia (agentic systems in banking, humanoid robots in logistics, AI labs running end-to-end scientific R&D). Geopolitics is the primary engine of AI-Dystopia this week, from TSMC's Arizona expansion to the 29-nation AI governance body formed without the US, to Xi's Shanghai speech. Regulation is carrying both AI-Dystopia signals (Australia's Office of AI, New York's data centre moratorium, EU versus Google) and some Futurama ones (Singapore's mandatory AI training notifications, DeepMind's push for a standards body). Trust is the dominant driver of Paperclipalypse – 50 of the 71 trust-driver signals landed there, from autonomous AI worms to Meta's alleged AI-driven discriminatory layoffs to hallucinated government policy reports. Talent is split: rising in Futurama (manager accountability for AI adoption, TCS hiring) and increasingly visible in Fizzle (AI literacy eroding human connection, Gen Z resistance, the vanishing path from junior to senior engineer when AI writes the code).
The scan is almost entirely silent on mid-market and SME AI adoption. The conversation is dominated by hyperscalers, frontier labs, government policy and large enterprises. The question of whether AI is actually working for a 50-person professional services firm, a regional logistics company or a New Zealand SME barely registers. That's a real gap, not just an editorial choice. If the benefits are concentrating at the top and the conversation is concentrating there too, smaller organisations are navigating this mostly blind.
If I were a CEO or CIO reading this scan, here's what I'd take to my leadership team. The power-pooling dynamic is structural, not cyclical. The vendors your organisation depends on – for compute, for models, for platforms – are actively consolidating their leverage over you. Nadella's "reverse information paradox" warning isn't abstract: every prompt your team sends to a frontier model is potentially training your competitor's next advantage. The Albanese government is trying to set rules before leverage is lost. You should be doing the same thing inside your own organisation. Map where your AI dependencies sit. Identify which vendors could lock you in. Ask whether your data is working for you or for them. The organisations that come out of this decade well won't be the ones who adopted AI fastest. They'll be the ones who adopted it with their eyes open.
The articles most responsible for bending the five worlds this week.
Australia's Prime Minister announced a new Office of AI to set national standards and oversee data centres – the biggest AI policy move in the country's history. It matters because it signals that even governments now see AI power pooling dangerously in a few hands, and they're trying to write the rules before it's too late to change anything.
Apple sued OpenAI, claiming former employees took secret information about Apple's AI plans and used it to help build ChatGPT. This is a big deal because it shows how the biggest tech companies are fighting over ideas, people and data – and ordinary businesses and workers are caught in the middle of that battle.
IBM lost $69 billion in market value in a single day after warning that companies are spending so much on AI hardware that they're cutting their regular software budgets. That's a sign that AI spending is growing faster than the real benefits, which is exactly what happens when the hype runs ahead of the proof.
Workers are suing Meta, claiming an AI system unfairly targeted employees on protected leave when deciding who to sack. Nobody can fully audit how the AI made those calls – and that's the core problem when powerful tools make life-changing decisions that humans can't easily check or reverse.
A New Zealand government adviser submitted a policy report full of fake scientific references that ChatGPT had invented, because he was rushed and trusted the tool without checking. Real government decisions could have been made on made-up evidence – a reminder that AI can cause genuine harm when people use it without the right guardrails in place.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free