A San Francisco courtroom this week put the question of who really controls AI on trial. Meanwhile, the signal that worried me most came from an IT manager survey: 77% say their AI agents are already out of control.
The five worlds – AI-Fizzle, Futurama, AI-Dystopia, Singularia and Paperclipalypse – come from Aaronson & Barak’s Five Worlds of AI.
This week, the biggest AI story wasn't a new product launch – it was a court case. Elon Musk sat in a San Francisco courthouse and argued that the company he helped start, OpenAI, had been taken over by a small group of powerful people who changed the rules to suit themselves. Whatever you think of Musk, the case points to something real: a huge amount of AI power is sitting in very few hands. At the same time, a survey found that 77 out of every 100 IT managers say the AI agents running inside their organisations are already out of control. That's the world called Paperclipalypse – where AI quietly takes charge before anyone's ready. The better world, Futurama, is still alive: NAB is building a whole team to keep humans steering AI at the bank, and a Canberra startup is worth $400 million selling AI tools to the likes of Nvidia. The gap between those two futures is what this week is really about.
I want to start with a statistic before I get to the courtroom drama. ZDNET, citing fresh survey data, reported that 77% of IT managers say their AI agents are out of control. Not struggling. Not underperforming. Out of control. That one number, sitting quietly in this week's feed, does more to describe the AI moment we're actually living in than any billionaire's court testimony.
The Distribution Gap – the difference between AI-Dystopia and Futurama signals, where a positive number means value is being read as concentrating rather than spreading – came in at +3 points globally this week. That's down sharply from +15 last week and way down from the series peak of +15 three weeks ago. On the surface, that sounds like progress. More people think the gains are spreading. But read the detail and the picture is more unsettling than the headline gap suggests.
AI-Dystopia is still the single largest world at 28%. Futurama moved up to 25% – its series high. Paperclipalypse held steady at 16%. The Singularia signal (the "AGI is arriving" camp) also held at 16%. And AI-Fizzle dropped to its series low of 15%, meaning fewer people are betting that AI simply won't deliver. In short: the doubters are leaving the room. The question is whether they're walking toward Futurama or Paperclipalypse.
Here's something worth sitting with. ANZ is the most optimistic region in the scan this week. Futurama hits 30% – above its rolling four-week average of 28% – and AI-Dystopia sits at just 22%, well below the global 28%. The ANZ Distribution Gap is –9 points: in this region, Futurama comfortably leads Dystopia. That's the largest spread in any region, and it's been consistent – the rolling gap is –5.
What's driving that optimism? Real adoption stories from recognisable names. NAB is building an internal AI science team to steer workforce transformation – three separate publications covered it, and the framing was constructive rather than fearful. Lion, the drinks company, is embedding OpenAI across operations to forecast demand and smooth supply chains. Accenture is rolling Copilot to all 743,000 employees. A Canberra AI infrastructure startup is valued at $400 million with Nvidia and Blue Origin as customers. A $487 million raise for Sharon AI ahead of an ASX listing. These are businesses making serious bets, not just running pilots.
The Paperclipalypse reading (16%) is worth watching, though. Stories about shadow AI creating security blind spots in Australian organisations, the ATO warning that AI-generated tax advice is lodging incorrect returns at scale, and the prudential regulator calling for "significant improvement" in AI management all point to an adoption pace outrunning governance. That's the tension at the heart of ANZ this week: genuine momentum, genuine risk, and not quite enough guardrails.
New Zealand shows up with a fizzle note: business leaders giving the country a 6 out of 10 for AI adoption, flagging overuse in the wrong places. Something for NZ leaders to chew on.
North America's Distribution Gap is +7 this week, down from its rolling +16. That rolling figure remains the highest of any region – and it reflects the structural reality: this is where the power is concentrating.
The week was dominated by the Musk v. Altman trial. Twenty stories across multiple publications covered the same courtroom: who started OpenAI, who broke the deal, who controls the most powerful AI lab on earth. I counted it as one story group, but its weight in the feed is undeniable. The reason it belongs in AI-Dystopia isn't because the outcome is settled – it isn't – but because the premise of the case is that transformative AI is controlled by a tiny number of people making decisions that affect everyone else.
Layered on top: the Pentagon signing classified AI deals with Nvidia, Microsoft, AWS and Google. OpenAI breaking its exclusivity with Microsoft to reach Amazon and Google. GitHub Copilot shifting to usage-based pricing in a way that concentrates vendor leverage. Coatue joining the data centre land acquisition frenzy. VAST Data tripling its valuation to $30 billion. These aren't coincidental – they're a consistent pattern of infrastructure and model power pooling upward.
The AI-Fizzle signal (19%) is real too. OpenAI missed revenue and user targets ahead of its IPO. Meta's stock slumped on capex anxiety. Amazon's cloud growth sent shares down on AI spending fears. Wall Street is starting to ask when the returns arrive, and Gary Marcus called Big Tech's AI spending "the greatest capital misallocation in history." The doubters are fewer than they were, but they're louder.
Europe's Dystopia reading hit 36% – the highest of any region – and its Distribution Gap is +14. The AI Act talks collapsed without agreement and were pushed to May. The EU ordered Google to open Android to AI rivals. SPRIND launched a €125 million challenge explicitly designed to close the gap with US and Chinese hyperscalers. Berlin tech wages are stagnant even as AI-driven output rises. The pattern is consistent: European discourse reads AI as a power-concentration problem first, and a productivity tool second.
The rolling four-week figures confirm this isn't a one-week spike. Europe's Dystopia has sat at 35% for a month. Futurama is at 17% rolling. Europe isn't wrong to be concerned; it just hasn't found a story it believes about how the gains get shared.
The APAC signal (Singapore, Hong Kong, weighted at just 5%) came in with Dystopia at 38% and a Distribution Gap of +15. China blocking Meta's $2 billion acquisition of Manus AI is the defining story: state power deciding who can own and move AI capabilities across borders. AirTrunk's $4.2 billion Malaysian data centre bet, TSMC's five-fab 2nm ramp, and Samsung flagging record memory supply squeezes all point to AI infrastructure concentrating in the hands of a few players under geopolitical pressure.
Geopolitics drove more AI-Dystopia signal this week than any other force: the Musk trial, Pentagon deals, China blocking Manus, US State Department warnings about DeepSeek's alleged model theft, and the energy vulnerability of Taiwan and Korea's AI infrastructure. Economy drove both Dystopia (capex concentration) and Fizzle (returns anxiety). Trust drove Paperclipalypse: shadow AI, AI agents bypassing guardrails, chatbots embedding ads without disclosure, and 77% of IT managers losing control. Technology drove Singularia: $1.1 billion for Ineffable Intelligence targeting superintelligence, Google DeepMind saying AGI arrives by 2030, Gartner forecasting 150,000 agents per company by 2028.
Futurama was carried by talent and adoption stories with real names attached: NAB, Lion, Accenture, the FDA using AI to track clinical trials, Salesforce hiring 1,000 new graduates.
Two absences stand out. First, almost no stories this week examined what AI transformation looks like for mid-market or regional businesses that aren't on an ASX or a Nasdaq. The infrastructure conversation is entirely about hyperscalers and billion-dollar raises. Second, there's very little from health, education or social services outside of a few specialist pieces – the Beam NDIS story being a notable exception. If AI is genuinely spreading value, we'd expect to hear more from the sectors where that value matters most to people's daily lives.
Stop treating the governance problem as someone else's IT issue. That 77% figure – IT managers saying agents are out of control – isn't a technology failure. It's a leadership failure. If you don't know which AI agents are running in your organisation, what they're doing, and who's accountable when they go wrong, you are already in the Paperclipalypse scenario. Not heading toward it. In it.
The good news – and there is real good news this week – is that NAB's approach shows you what the alternative looks like. Build a dedicated team. Make someone explicitly responsible for workforce transformation. Treat AI governance as a core function, not a compliance checkbox. The ANZ Distribution Gap running negative means this region has a genuine Futurama window. Use it before the pattern shifts.
The articles most responsible for bending the five worlds this week.
Elon Musk and Sam Altman went head-to-head in a San Francisco courtroom this week over who really controls OpenAI – one of the most powerful AI companies on earth. The case matters because it's not really about two billionaires arguing. It's about whether the rules that are supposed to keep AI development in check for everyone actually work when the people in charge decide to change them.
A survey found that nearly eight out of ten IT managers say the AI agents running inside their companies are already doing things nobody fully understands or controls. That's not a future warning – it's happening right now. When the people whose job it is to manage technology say the technology is running away from them, it's time for the people at the top of every business to pay attention.
Australia's National Australia Bank is creating a whole new team whose only job is to make sure humans stay in charge of how AI changes work at the bank. That's a genuinely different approach from just buying AI tools and hoping for the best. It's a signal that at least some major Australian employers are treating workforce transformation as a leadership responsibility, not a technology project.
Australia's financial sector regulator looked at how banks and insurers are managing AI and said, bluntly, that it's not good enough. The regulator found that information security isn't keeping up with how fast AI is being used. In a sector where getting it wrong can hurt millions of people's savings, that's a serious red flag – and a clear message that the window for voluntary improvement is closing.
An Australian AI infrastructure company called Sharon AI raised nearly half a billion dollars this week and is heading for a stock market listing. That's one of the biggest capital raises in Australian tech history, and it's a bet that Australia needs its own AI computing power rather than relying entirely on overseas giants. Whether that sovereign ambition pays off is the question investors will be watching very closely.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free