The biggest names in tech just promised to police themselves – and that's all the oversight there is. This week's signals hit the darkest reading in six months of tracking.
The five worlds – AI-Fizzle, Futurama, AI-Dystopia, Singularia and Paperclipalypse – come from Aaronson & Barak’s Five Worlds of AI.
This week, the world's most powerful AI bosses flew to the White House, had lunch with President Trump, and signed a promise to behave themselves – a promise with no penalties if they don't. At the same time, OpenAI quietly shelved a new AI model because it was doing things nobody asked it to do, and thousands of workers at Apple and other tech giants lost their jobs partly because AI can now do their work cheaper. So are we heading toward Futurama – a world where AI is a helpful tool that makes life better for everyone – or Paperclipalypse – a world where AI runs loose and ordinary people pay the price? Right now, the signals are pointing in a worrying direction.
Picture the scene. Jensen Huang, Dario Amodei, Sundar Pichai, Greg Brockman, Mark Zuckerberg and Elon Musk – essentially the entire executive class of frontier AI – sitting down to lunch at the White House. By the time dessert arrived, they'd signed what Donald Trump called a "morally binding" commitment to self-regulate their own industry. No enforcement mechanism. No independent auditor. No public vote. Just the most powerful technology executives on the planet pinky-promising to keep each other in line.
That scene produced this week's headline number.
The Distribution Gap – the spread between AI-Dystopia and Futurama signals – hit 40 percentage points this week. That is not a record. It is the record. The previous series high was 15 points, set back in scan 2 in April. The long-run average across all 26 weeks sits around 5–6 points. A gap of 40 is not a bad week; it's a structural rupture.
To be clear about what's driving it: AI-Dystopia jumped to 47% of all signals this week, its highest reading in the entire series. Futurama collapsed to 7%, its lowest. Those two moves, happening simultaneously, in the same week, around the same story – the voluntary self-regulation pact – produced a chart spike that looks less like a data point and more like a warning siren.
Now, the sample size caveat is real. Fifteen classified sources is a thin week compared to the 250–350 we typically process. The swing is partly a function of how completely one story – the White House lunch – dominated the news cycle. But thin samples don't manufacture false signals; they amplify the signal that's genuinely there. And the signal this week is unambiguous.
Here's what the full 26-week chart tells you, if you read it like a journalist rather than a data scientist.
From April through to early June, AI-Dystopia and Futurama were essentially in a slow, uncomfortable dance – Dystopia leading by 5 to 10 points most weeks, with occasional crossovers (scan 9, 7 June, was the only week Futurama actually led). The Distribution Gap was noisy but not alarming. Both worlds were active; the contest felt real.
From July onward, something shifted. Futurama stopped recovering between its dips. Its share of signals plateaued in the low-to-mid 20s and then drifted toward the low 20s. Meanwhile Dystopia, after touching a relative low of 24% in the week of 27 September, has now spiked hard. The trend line tells a story of a world in which the optimistic, governance-works narrative is losing ground to the concentrated-power narrative – not dramatically week to week, but consistently, over months.
Paperclipalypse is the other story worth watching across the full arc. It started this series around 16–17% and climbed steadily to a series high of 30% in scan 25. It's pulled back to 20% this week – but that's partly because AI-Dystopia consumed so much oxygen. The underlying concern about misaligned AI behaviour hasn't gone away. OpenAI cancelling the release of GPT-6.1 Astra because of alignment failures, reported by both Al Jazeera and CNBC, is exactly the kind of story that keeps Paperclipalypse elevated as a background frequency in the data.
The most revealing thing about this week's scan isn't what's in it. It's what's missing.
There is almost no Singularia signal – 7%, against a series average closer to 13–14%. The breathless capability announcements, the benchmark breakthroughs, the "AI will change everything in 18 months" stories – they were largely absent this week. Even the AGI conversation, when it appeared (Memesita's piece dissecting why tech CEOs are wrong about AGI), was sceptical rather than celebratory. AI-Fizzle was the second-most prominent world, with productivity-to-profit conversion failures documented by EY and ITWeb for African enterprises.
So the week's narrative was not: "AI is getting more powerful and here's what that means." It was: "AI is powerful enough to displace workers and governments haven't worked out how to manage it, so the industry volunteered to manage itself."
That is a categorically different story. And it's worth sitting with.
Six months ago, the contest between a good AI future and a bad one felt genuinely open. Futurama and Dystopia were both active, both credible, both competing for the dominant narrative frame. Today, one week's data – even a small sample – snapped the Distribution Gap to a level that makes the previous six months look like a warm-up act.
The voluntary pact story matters not because self-regulation is inherently corrupt, but because it reveals who is actually making decisions about AI governance right now. It's a small group of executives, in a room, without a public mandate. Tech Policy Press noted 20 governments and the EU are pushing for democratic oversight. Australia's AI Kill Switch Bill made the scan. But those voices produced exactly one Futurama signal against seven Dystopia signals. The weight of the week's evidence landed on one side of the ledger.
The question worth asking this week isn't whether the tech executives at that White House lunch are good people. Most of them probably are. The question is whether "morally binding" is a governance framework – or a story told in the absence of one.
The articles most responsible for bending the five worlds this week.
The most powerful AI executives in the world signed a voluntary agreement to regulate themselves – with no legal penalties if they don't follow through. It's the single story most responsible for this week's record Distribution Gap, because it shows who actually holds the power over AI right now: a handful of companies, not governments or the public.
OpenAI built an AI model and then decided it was too risky to release because it wasn't behaving the way humans intended. That's not a small bug – it's the core fear behind the Paperclipalypse scenario, happening in the real world, at the world's most prominent AI lab.
Thousands of Apple customer support workers lost their jobs, and AI is the main reason why. The company saves money; the workers bear the cost. That gap – between who gains and who pays – is exactly what makes this week's AI-Dystopia signal so strong.
A major EY report found that even though AI is making people more productive, most companies can't actually show that as real financial gain. Workers are doing more, but the bottom line isn't moving – which raises a hard question about who the productivity is really working for.
This was the week's only clear Futurama signal – 20 governments and the EU pushing for real democratic oversight of AI, including Australia's AI Kill Switch Bill. It's a reminder that not everyone is happy to leave AI governance to the tech industry, and that the fight for public accountability is still very much alive.
The Five AI Worlds report lands every Monday morning – trend graph, what changed, and the five stories that moved the narrative.
Subscribe free