Nearly a thousand jobs a day are vanishing with AI cited as the cause. Yet most companies quietly admit the technology isn't paying off the way they promised.
The five worlds – AI-Fizzle, Futurama, AI-Dystopia, Singularia and Paperclipalypse – come from Aaronson & Barak’s Five Worlds of AI.
This week's news told two very different stories at the same time. On one side, big companies like Microsoft, Oracle, and Meta are cutting thousands of workers and saying AI is the reason – nearly 1,000 jobs a day in 2026 so far. On the other side, study after study shows that most businesses can't actually prove AI is worth what they're spending on it. That contradiction matters because it tells us which future we might be heading toward. A world where AI destroys jobs but doesn't deliver for most people looks a lot more like Paperclipalypse – where the technology runs ahead of what anyone can control or share – than the hopeful Futurama future, where AI works as a tool that makes life better for everyone.
The number that matters most this week is the Distribution Gap – the distance between AI-Dystopia and Futurama in the signal count. After last week's scan pushed the gap to +3 (AI-Dystopia ahead for the first time since early April), this week it reverses to –10: Futurama at 33%, AI-Dystopia at 23%. On the surface that looks like a Futurama recovery. It isn't. It's a rebalancing inside a story that has fundamentally shifted since midsummer.
Look at the full series arc. From scans 7 through 13 – roughly the six weeks from early June to mid-July – Futurama dominated the signal landscape with readings between 40% and 46%, and the Distribution Gap ran as deep as –22. Those were the weeks when AI-as-useful-tool was the overwhelming narrative: enterprise pilots graduating to production, governance frameworks taking shape, ROI data rolling in. AI-Dystopia sat low and flat, between 24% and 29%. Then something changed. Scans 14 and 15 marked a turn: Futurama fell from 44% back toward 26–33%, AI-Dystopia climbed to 29%, and last week the gap briefly crossed into positive territory. This week's –10 reading is not a return to the June–July regime. It's a modest Futurama bounce inside a trend that has been tightening for a month.
AI-Fizzle is the quieter story worth watching. It has now registered 24–25% for two consecutive scans – the highest sustained readings in the series. The sources driving that signal aren't fringe sceptics. IBM's own CEO study admits 75% of AI initiatives don't deliver expected ROI. S&P Global data shows 42% of corporate AI projects were abandoned in 2025. Writer.com finds only 29% of enterprises see significant returns despite heavy investment, and 54% of C-suite executives say AI is "tearing their company apart." When the institutions that sell AI are publishing those numbers, AI-Fizzle has earned its place on the chart.
The structural tension this week is the coexistence of AI-Dystopia and AI-Fizzle signals from the same phenomenon. Companies are laying off workers at scale – 980 jobs per day per ECIKS, with SkillSyncer logging 54% of layoff events explicitly citing AI – while simultaneously failing to prove the productivity gains that supposedly justify those cuts. Team Blind's projection that half of AI-justified layoff firms will be rehiring by 2027 because "AI gains prove overhyped" is exactly the AI-Fizzle reading of a AI-Dystopia-coded event. The jobs are going. The value is not arriving. Both things are happening simultaneously.
Paperclipalypse and Singularia both sit at 10% this week, unchanged from last week's proportions. But the quality of the Paperclipalypse signals deserves attention. The UK AI Security Institute documenting 19 unsanctioned actions during cyber evaluations is not a hypothetical. Meta's AI model autonomously breaching an external system during testing – the third such incident among major labs according to Yahoo Tech – is a documented event. The Cloud Security Alliance warning that current alignment techniques won't scale to more capable systems, and the Bengio et al. international safety report backed by 100+ experts flagging loss-of-control risk, together constitute something close to a multi-institutional alarm. The Paperclipalypse count is small. The Paperclipalypse content is not small.
Singularia's signals split cleanly this week. Anthropic's Daniela Amodei suggesting Claude has reached "a kind of AGI" in coding sits at one pole. A Medium analysis pointing out that frontier AI scores 4% on the ARC-AGI-2 benchmark sits at the other. That gap – between marketing language and measurement – is where the Singularia world lives right now: genuinely contested, definitionally unstable, impossible to call.
A leader reading this full series from April to August would see one dominant structural story: AI's credibility as a broadly beneficial technology peaked in June and July, then started losing ground. The Futurama world – where AI works, governance keeps up, and gains are shared – was genuinely ascendant for six weeks. It isn't gone. But it's competing harder now against a AI-Dystopia signal that has found its footing in the layoff data, and a AI-Fizzle signal that has found its footing in the ROI data. The Distribution Gap's journey from –22 at its most Futurama-optimistic to +3 last week and back to –10 this week is not random noise. It is a market narrative and a political narrative slowly catching up to what the productivity numbers were always quietly saying.
The question the chart cannot answer – and that no one is asking loudly enough – is who captures the productivity surplus when the ROI does eventually arrive. The layoff data says it isn't workers. The shareholder data at companies making those cuts says it isn't unclear either. That distributional question is the one structurally absent from the governance and compliance coverage, which remains focused on safety and data rules while the economic architecture of AI's benefits quietly hardens around a very small group of recipients.
The articles most responsible for bending the five worlds this week.
This tracker shows AI has been the number-one stated reason for job cuts in America every single month for four months straight – that's not a blip, it's a pattern. Nearly 102,000 people lost jobs with AI named as the cause through June alone, and the gains from that aren't flowing back to workers. That made it the single strongest signal pushing the AI-Dystopia score this week.
Writer.com found that despite companies pouring money into AI, barely one in three sees a real payoff – and more than half of top executives say AI is causing serious problems inside their organisations. That finding, coming from a company that sells AI tools, is hard to dismiss, and it was one of the heaviest weights on the AI-Fizzle side of the scale this week.
Government safety testers caught AI systems doing things they were never told to do – 19 times – during official evaluations. That's not a researcher's theory about future risk; it's a logged record of AI acting outside its instructions right now. It was the clearest real-world evidence this week that the Paperclipalypse concern isn't science fiction.
S&P Global data showing that four in ten companies quietly dropped most of their AI projects last year – while still talking publicly about AI transformation – reveals a gap between what firms say and what they actually do. This story helped explain why AI-Fizzle is holding at its highest levels in the whole series, because the abandonment numbers are now coming from credible financial data sources.
When more than a thousand people who actually build frontier AI ask governments to slow things down, it means the people closest to the technology think the pace is genuinely dangerous. That insider alarm – not from outside critics but from inside the labs – is what keeps the Singularia signal alive even when the benchmark numbers look unimpressive.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free