A cybersecurity alarm from a single AI lab rattled governments and banks worldwide this week. Meanwhile the data on who's actually winning from AI – and who isn't – got harder to ignore.
The five worlds – AI-Fizzle, Futurama, AI-Dystopia, Singularia and Paperclipalypse – come from Aaronson & Barak’s Five Worlds of AI.
This week, a new AI tool called Mythos from the company Anthropic spooked governments, banks and security experts around the world because it can find weaknesses in computer systems faster than humans can fix them. At the same time, new research showed that most of the money being made from AI is going to a small number of companies and workers at the top – not spreading to everyone. That's the core tension right now: AI is getting more powerful every week, but the benefits aren't being shared evenly. Think of two possible futures. In the hopeful one – Futurama – AI stays a useful tool and the good stuff flows to everyone. In the scary one – Paperclipalypse – AI gets so capable it causes harm people can't stop or control. This week, the signals pointed firmly toward the scary end of that scale.
Let me give you a number to start with: 14 points.
That's the global Distribution Gap this week – the difference between the share of signals pointing toward a world where AI's gains concentrate versus the share saying those gains spread. It was 10 points in our first scan a fortnight ago. Now it's 14 and rising. That's not noise. That's a trend finding its legs.
For context: a neutral reading is zero. Positive means value is being read as concentrating – in fewer companies, fewer countries, fewer workers. Negative means it's spreading. We're moving in the wrong direction, and the evidence this week was hard to look away from.
PwC research covered by Silicon Republic found 75% of AI's economic value is captured by just 20% of companies. MIT's 2026 jobs and work report showed power users pulling away from everyone else. Stanford data, picked up by TechCrunch, highlighted a widening disconnect between AI insiders and the broader public. The Wall Street Journal ran a CEO piece about straight talk on job losses. The Conversation published employment data showing early signs of structural disruption hitting white-collar roles.
None of these are outliers. They're a pattern.
Here's what's genuinely interesting about the ANZ picture this week: it doesn't look like the global picture.
The regional Distribution Gap is zero for the week – dead flat, compared with the rolling four-week figure of +3. That matters. It means local voices are still debating whether AI's gains concentrate or spread, while North America (+20 this week) and Europe (+32, a sharp spike) have largely stopped asking the question.
What drove that relative balance? Westpac hiring a Chief AI Innovation Officer – a real institutional commitment, not just a press release. Westpac NZ deploying AI to augment contact centre staff rather than replace them. The Federal Court formally embracing AI with a practice note that keeps humans setting the rules. TPG Telecom already cutting fault resolution times from hours to minutes. WhiteHawk acquiring Quixxi to build out AI governance tools. Essential Energy turning AI on safety inspections. These are practical, human-led stories.
But the counterweight is real too. Capital Brief's KPMG survey found Australia lags peer economies on AI-driven productivity. The AFR ran a blunt piece: AI is moving fast, most organisations aren't. Entry-level jobs disappearing for young workers. Public servants being cognitively deskilled. Nine instructing journalists to increase AI use. The Fair Work Commission being swamped by AI-assisted meritless claims.
The rolling Futurama share for ANZ (24%) is actually higher than North America's (16% rolling). That's a meaningful signal. Local adoption stories tend to be more measured, more people-centred. The challenge is translating that culture into productivity gains before the gap with the US widens further.
The North American Distribution Gap of 20 points this week – consistent with the 19-point rolling figure – reflects a market that has largely resolved the question in favour of concentration. Snap cut 16% of its workforce while increasing AI investment. PwC's leadership sent a clear message to staff: embrace AI or become irrelevant. OpenAI is acquiring startups, expanding into life sciences, pushing Codex to take control of the desktop, and in talks for a US$40 billion loan. TSMC posted a 58% profit surge. ASML lifted its 2026 forecast.
The Allbirds story is almost too perfect as a metaphor. A Kiwi-founded shoe brand ditches its product entirely, slaps AI on the rebrand, watches its share price surge 700% then crash 36% the next day. AI-washing confirmed, hype cycle visible in real time.
Europe's Distribution Gap hit 32 points this week, up from a rolling 18. That's a significant single-week move, though European signals are fewer (31 stories) so individual stories swing the reading harder. The EU ordering Meta to change its WhatsApp AI policy, OpenAI's Stargate UK retreat exposing sovereign dependency, Sifted's analysis that AI is swallowing half of all European tech funding, ASML's booming order book – these all point the same way. Concentration of power, infrastructure and capital in a handful of players, with governments scrambling to build sovereign alternatives that are almost comically underfunded relative to the hyperscalers.
APAC at a +13 Distribution Gap is actually the most optimistic of the major regions this week, though that's partly Singapore's balance – IHH Healthcare deploying AI across Mount Elizabeth with efficiency gains reaching patients, Singapore's exports up 15.3% on AI electronics demand, the digital minister talking AI agents for inclusion. Malaysia securing US$22 billion in digital investment. China's daily AI token usage hitting 140 trillion and climbing.
The shadow side: ByteDance and Tencent in an aggressive talent war, TSMC's 58% profit surge reflecting where chip value is pooling, data centre booms reshaping Batam and Johor with infrastructure gains going to large operators.
The most-covered story of the week – 18 separate pieces – was Anthropic's Claude Mythos model and Project Glasswing. This deserves plain language.
Anthropics' tool can find software vulnerabilities at a speed and scale that breaks the existing cybersecurity playbook. The US White House gave agencies access to it. Australia was warned it's exposed because it doesn't have that access. Goldman Sachs partnered with Anthropic specifically to assess the risk to banks. Super funds sought ACCC approval to collaborate on cyber defences before the model's capabilities become widely available. Singapore urged firms to shore up defences. A trade group warned of systemic financial market disruption.
This is Paperclipalypse territory – not because AI has gone rogue, but because the capability exists and the containment doesn't. That's the definition of the control problem in practice.
Technology is the dominant force (120 signals) – frontier model releases, agentic AI arriving in enterprise infrastructure, coding tools gaining desktop control. Trust is the second force (73 signals), almost entirely in the Paperclipalypse world: cybersecurity harms, AI chats used as legal evidence, mortgage fraud harder to detect, workplace surveillance, model reliability failures. Economy (67 signals) splits between Dystopia and Fizzle – capital concentrating at the top of the AI supply chain, ROI hard to prove for most adopters. Talent (44 signals) sits almost entirely in Dystopia territory: entry-level jobs disappearing, AI fluency becoming a hiring gate, talent debt accumulating invisibly. Geopolitics (31 signals) is pure Dystopia – US-China chip war, sovereign AI programmes, export controls and their unintended consequences.
Notably absent from the discourse: serious coverage of AI's environmental costs at the enterprise level (water and energy appear but aren't driving strategic decisions), meaningful stories about AI benefits flowing to customers rather than shareholders, and almost nothing about mid-market or small business adoption in ANZ beyond vendor events.
Here's the plain reading for the CEO or CIO in the room.
The AI productivity gains are real. The ANZ stories prove it – TPG, Westpac, Essential Energy, the Federal Court. But the KPMG data says Australia is lagging peers, and the Forrester research says employees aren't ready. You can't close that gap with another pilot programme. You close it by picking one workflow, fully redesigning it around AI, measuring the outcome and being honest about what happened to the people in it.
The Mythos/Glasswing story is not something you can wait on. If your CISO hasn't briefed the board on AI-enabled attack surfaces in the last 90 days, that briefing needs to be scheduled this week. The Australian government's cyber exposure is your cyber exposure.
And if your AI strategy is mostly about cost reduction, take a hard look at what you're building. The Distribution Gap is widening because the value is flowing up, not out. The organisations that will win the next phase are the ones that make AI genuinely useful for the people doing the work – not just the people watching the numbers.
The articles most responsible for bending the five worlds this week.
Anthropic's new AI tool can automatically find weaknesses in computer systems faster than any human or existing security tool can respond. Governments, banks and critical infrastructure providers around the world were alarmed – and Australia was warned it's exposed because it lacks access to the same defensive capability. This was the defining story of the week.
A KPMG survey found Australian businesses are much more cautious about AI than their counterparts in the US and Canada, and that caution is showing up as a real productivity gap. This isn't just about being careful – it means Australian organisations are missing out on gains their competitors are already capturing.
Entry-level jobs are disappearing as companies use AI to do work that used to give young people their first foothold in a career. The savings go to companies and shareholders, not to the workers being replaced. It's one of the clearest signs that AI's gains are not being shared evenly.
PwC research found that three quarters of AI's economic value is flowing to just one in five companies. That means most organisations are investing in AI but the returns are pooling at the top. It's the most concrete single number explaining why the gap between AI winners and everyone else keeps getting wider.
Westpac creating a dedicated Chief AI Innovation Officer is a real signal of institutional commitment – someone at the leadership table whose job is to make sure AI works for the whole organisation, not just the tech team. In a week dominated by alarm and concentration stories, this was a genuine piece of good news from Australia.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free