For the first time in nine weeks, the global signal flipped: AI looks more like a shared benefit than a power grab. Don't celebrate yet – the costs are quietly shifting onto workers, households and governments.
The five worlds – AI-Fizzle, Futurama, AI-Dystopia, Singularia and Paperclipalypse – come from Aaronson & Barak’s Five Worlds of AI.
This week, the big news is that more stories worldwide described AI as something that could help lots of people – not just a handful of powerful companies – than any week we've tracked so far. That's the Futurama future: AI as a useful tool, humans still in charge, benefits spread around. But there's a catch. Even as the headlines got more hopeful, the costs of AI – soaring power bills, job cuts, runaway spending on computer chips – kept landing on ordinary people and smaller businesses. That's the shadow of Paperclipalypse: a world where AI causes harm that nobody planned for and nobody can quite stop. So this week is genuinely good news, but it's complicated good news.
Here's something worth stopping for. For the first time in nine scans – since I started tracking this in April – the global Distribution Gap has gone negative.
The Distribution Gap measures how much more AI-Dystopia coverage there is than Futurama coverage. When it's positive, the dominant signal is that AI's gains are pooling at the top: with hyperscalers, frontier labs and a handful of sovereign actors. When it's negative, more of the discourse says AI is spreading value broadly – augmenting people, not replacing them.
This week: –4 points. Futurama at 27%, AI-Dystopia at 23%.
That's a reversal. Last week the gap was +1. The series high was +15 in the second scan. The trend from scans 6 through 8 was a slow drift toward equilibrium. Scan 9 breaks past it.
Is this a turning point? Or noise? My honest read: continuation of a real trend, but not a declaration of victory. The rolling four-week global picture still shows Dystopia outpacing Futurama in North America and APAC. And the stories driving the Futurama lift this week include a lot of industry voices – vendors, announcements, optimistic CEOs – which I always separate from genuine adoption evidence. The signal is real but it needs company.
Australia and NZ had a big week for Futurama signals – 36%, the highest of any region, and well above the 32% rolling four-week average. The Distribution Gap here is –21, meaning Futurama outpaces Dystopia by more than three to one.
That's genuinely striking. What's behind it? Government agencies publishing AI frameworks (DEWR, ASIC, the NZ government's regulatory guidance), the judiciary proactively engaging with AI, HBF deploying its first AI agent for members, Telstra's CEO articulating a clear AI strategy, PwC pushing 80,000 staff from AI access to actual use. These are adoption stories, not just announcements. People in organisations making deliberate choices.
But here's the honest complication: ANZ's Paperclipalypse reading is also elevated this week at 23%, above its rolling average of 19%. The Anthropic Mythos story – an AI model deemed too dangerous for general release, now handed to Australia's national intelligence agency and 150 critical infrastructure organisations – drove most of that. The Australian Financial Review, CIO, The Conversation, Cyber Daily and SmartCompany all covered it. That's a story with two legitimate readings at once: cyber resilience improving, and a powerful capability deployed into critical systems before anyone's quite sure it's safe.
The AI-Fizzle reading in ANZ is also persistent at 22%, above rolling average. Cost concerns from CBA, readiness gaps from multiple HR publications, and the NZ government's Budget barely mentioning AI all contributed. The training gap is real. Atlassian's VP of HR told HRD Australia that AI is making some teams slower, not faster. That's worth sitting with.
North America tells a different story. AI-Dystopia at 28% this week, Singularia at 22%. The Distribution Gap is +9 – positive, meaning concentration signals still dominate.
The Anthropic IPO filing at a near-trillion-dollar valuation, Alphabet raising $80 billion for AI infrastructure, Trump signing an executive order giving the US government privileged pre-launch access to frontier models, and the AI-equity-stake debate between Sanders and the White House – these dominated the North American signal. Capital and sovereign power are concentrating in AI at an extraordinary pace.
The Singularia reading – autonomous AI taking over complex workflows – is the highest of any region this week, driven by Microsoft Build (MAI-Thinking-1, Scout, the GitHub agent strategy), Meta's Zuckerberg declaring AI agents will run entire businesses, and Snowflake's full pivot to the "Agentic Enterprise." These are industry stories, and I flag that. But when Microsoft, Meta, Snowflake and GitHub all announce the same thing in the same week, the direction of travel is clear.
Token costs are also front and centre. The TechCrunch, The AI Daily Brief and Business Insider coverage of "tokenmaxxing" and runaway AI spend is the clearest Fizzle signal of the week: spend is real, value is disputed, and Goldman Sachs' own AI sceptic is asking publicly when AI investment must convert to actual revenue.
Europe sits at a clean equilibrium this week – Futurama and Dystopia both at 26%, Distribution Gap at zero. The rolling four-week gap is +3, so this is a slight Futurama lean versus trend.
The EU sovereignty package, the UK regulator ordering Google to allow publisher opt-outs from AI search, the Morgan Stanley forecast of 20% banking job cuts, and the Helsing defence AI raise all pulled toward Dystopia. The Wordsmith legaltech raise, the Isembard industrial AI funding, and the Merantix early-stage fund pulled toward Futurama. Europe is genuinely split between building its own AI future and watching the concentration dynamics of the US and China with justified concern.
APAC (Singapore and Hong Kong signals, weighted at 5%) shows AI-Dystopia at 49% this week, the highest Dystopia reading of any region. DeepSeek's $7 billion raise, the US-China chip export story, Alibaba's elite committee reorganisation, TSMC's supply warnings and the sovereign wealth fund infrastructure bets all pointed the same direction: AI capability concentrating at the state and platform level in Asia, fast.
Breaking the week down by macro force makes the picture sharper.
Technology is the dominant driver – 115 signals – sitting behind both Futurama (new enterprise deployments, agent tooling, industrial AI) and Singularia (frontier model releases, autonomous coding, agentic infrastructure). The tech is genuinely moving.
Economy drives most of the Dystopia signal: hyperscaler capex races, IPO valuations, AI infrastructure cost externalisation onto households and grids, and PE capital concentrating in AI assets.
Trust underpins almost all of Paperclipalypse this week – 54 signals. The Anthropic Mythos story, the Florida lawsuit against OpenAI, AI-generated lawsuits flooding courts, self-replicating AI worms, deepfake CEO impersonation, and Anthropic's own call for a global development pause all sit here. The trust deficit is structural, not episodic.
Geopolitics is smaller in volume but potent: the US-China chip war, Canada and the EU's sovereign AI strategies, and the Trump order giving government privileged model access all reflect AI being reshaped by national power competition.
Talent drives a persistent Fizzle signal: workforce readiness gaps, graduates treating AI as cheating, the BCG finding that workers save a day a week with AI but don't know what to do with the time, and Atlassian's HR VP on teams getting slower.
The discourse this week said almost nothing about mid-market and SME adoption pathways. The stories were hyperscalers or startups, frontier labs or government agencies. The question of how a 200-person professional services firm in Brisbane or a 50-person manufacturer in Christchurch actually captures AI value – without a Microsoft enterprise agreement and a six-figure data team – was nearly invisible. The Business Times flagged it from Singapore, but ANZ was quiet.
There was also very little on productivity measurement. We have BCG telling us AI saves a day a week but nobody knows what to do with it. We have CBA warning businesses will tighten AI spend. We have Goldman's AI sceptic asking when the money converts to revenue. But almost no one published a rigorous framework for how a CFO should actually measure AI's contribution to the P&L this week.
If you're a CEO or CIO, the Distribution Gap flipping is legitimately good news – it means the broader discourse is starting to reflect what the smartest operators have been building toward: AI as a tool that spreads value, not just a technology that concentrates it.
But here's what I'd do on Monday. Stop measuring AI activity (subscriptions, tokens, tools deployed) and start measuring AI outcomes – decisions improved, costs avoided, revenue attributable. The tokenmaxxing story is a warning shot. Boards are starting to ask the Goldman question: when does this convert to money? You want your answer ready before they do.
The articles most responsible for bending the five worlds this week.
Anthropic built an AI that's so good at finding software weaknesses that they refused to release it to the public. This week, Australia's national intelligence agency and hundreds of critical infrastructure companies got access to it anyway. It might make the country safer from cyberattacks – but it also puts an extremely powerful, potentially risky tool into systems that millions of Australians depend on, like hospitals and power grids. That tension made it the most-discussed AI story in Australia all week.
Anthropic, the company that also made the Mythos AI model, filed paperwork this week to list on the stock market at a valuation near one trillion dollars. That's an almost unimaginable amount of money for a company that's only a few years old. It shows how much wealth is piling up in just a handful of AI companies – and raises a fair question about who really benefits from the AI boom.
Google's parent company announced plans to raise $80 billion – that's eighty thousand million dollars – to build more AI infrastructure. Only a tiny number of companies in the world can spend at this scale. When the cost of competing in AI gets this high, it means smaller companies, smaller countries and ordinary workers get further and further from the action.
In the same week it filed to raise billions on the stock market, Anthropic published a warning saying AI might soon be able to improve itself without human involvement – and asked the world to slow down development until we're sure it's safe. It's a remarkable thing for an AI company to say about its own technology. It was the clearest signal this week that even the people building AI are genuinely worried about where it's heading.
The AFR made the case this week that AI could let brave CEOs do things their companies used to pay expensive consultants to do – giving organisations more independence and more capability at lower cost. It's a hopeful story about AI spreading power to more people, not concentrating it. And it matters because it came from Australia's leading business paper, aimed squarely at the leaders who need to hear it.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free