14 September 2026
Current assessment: AI is changing capitalism faster than it is replacing it. Concentrated AI capitalism is the strongest observable trajectory.
The evidence does not currently support either of the two most dramatic claims: that AI is about to end capitalism, or that present-day AI is about to destroy humanity. What it does support is a less theatrical but potentially more important transition. AI capability, investment and adoption are rising quickly; ownership of frontier development remains concentrated; productivity gains are appearing in some forms of work; and the first labour-market effects are showing up most clearly in hiring pipelines and among younger workers.
Current trajectory table
| Possible future | Direction now | Confidence in direction | TWIS reading |
|---|---|---|---|
| AI-enhanced capitalism | ↑ Strengthening | High | The clearest present path: firms use AI to raise output and reorganise work. |
| Concentrated AI capitalism | ↑ Strengthening | High | Frontier models, compute and capital remain concentrated, so ownership gains can compound. |
| Managed/social capitalism | → Unclear | Low | Redistribution and social-policy responses exist as possibilities, but they are not yet the dominant response. |
| Post-work transition | ↗ Early signal | Low–Medium | Entry-level disruption is visible, but economy-wide displacement is not. |
| Authoritarian AI society | ? Watch | Low | AI can strengthen surveillance and control, but this tracker does not yet have enough evidence to call it the leading trajectory. |
| Abundance economy | ↑ Strengthening | Medium | AI is creating measurable consumer value and productivity gains, although those gains are uneven. |
| Major AI catastrophe | → No directional signal | Low | Serious misuse and systemic risks exist, but there is no evidence that catastrophe is becoming the realised economic trajectory. |
| Loss of human control | ? Watch | Low | Relevant capabilities are improving, but current systems still lack the sustained autonomous capability required for genuine loss of control. |
What has actually happened so far
The strongest labour evidence remains mixed rather than apocalyptic. Cross-country occupational research and payroll evidence still point more clearly to job transformation, changing hiring patterns and selective displacement than to economy-wide elimination of human work. Exposure is broad, but exposure is not the same thing as redundancy, and aggregate labour-market evidence has not yet shown a general collapse in employment.
There is nevertheless a warning signal at the entry point. Recent payroll research finds a measurable employment gap for younger workers in highly AI-exposed occupations, while more experienced workers have not shown the same pattern. The adjustment appears more through weaker hiring than a wave of dismissals. That matters because a labour transition can begin by closing entry routes before it becomes visible as mass unemployment.
Ownership may matter more than unemployment
The second signal is stronger. Frontier AI remains capital-intensive and highly concentrated. Model development, compute and infrastructure are still dominated by a relatively small set of large firms and capital providers. UK Government foresight likewise treats ownership and access as a critical uncertainty and warns that the gains can accrue disproportionately to frontier firms, capital owners and those controlling critical inputs.
This means society does not need mass unemployment for AI to increase inequality. If AI makes a worker more productive but the extra value is captured mainly by the owner of the model, compute infrastructure or firm, total output can rise while labour’s bargaining position weakens. Economic modelling also points to a distinction between wages and wealth: even where AI changes wage inequality in complicated ways, higher returns to capital can increase wealth concentration.
For this tracker, that is why concentrated AI capitalism currently ranks above post-capitalism. The machinery for private ownership already exists. A new social settlement does not.
The abundance signal is real too
Concentration is not the whole story. Research is also finding measurable consumer value and productivity gains in some structured forms of work. AI tools remain free or inexpensive for many users compared with the value they can provide. If capability continues improving while inference, robotics, energy and scientific discovery become cheaper, an abundance economy becomes more plausible.
But abundance and inequality are not opposites. They can occur together. A society can become capable of producing much more while access to the productive assets remains concentrated. The political question is therefore not simply whether AI makes society richer. It is who owns the systems that create the additional wealth, and how the resulting purchasing power is distributed.
Humanity is not currently facing a demonstrated loss-of-control event
International safety assessments draw an important boundary: current systems do not yet demonstrate the combined, sustained capabilities required for genuine loss of control, although capabilities related to autonomy, planning and undermining oversight are improving. A true loss-of-control scenario would require several abilities to work together reliably over long periods in real environments, including evading oversight, executing long-term plans and resisting countermeasures.
That does not make the risk imaginary. It makes it uncertain and forward-looking, rather than a description of what is happening now. The tracker will therefore keep loss of control visible without allowing it to dominate evidence about labour, ownership and political economy that is already observable.
The September 2026 baseline
The baseline judgement is therefore:
AI-enhanced capitalism is happening. Concentration is strengthening. Early labour displacement is visible but not yet economy-wide. Abundance is becoming technically more plausible. A managed redistribution settlement has not yet caught up. Catastrophic misuse and loss of control remain serious risks to monitor, but they are not the strongest description of the world we can currently observe.
The most important contest may not be humans against AI. It may be the contest over ownership of increasingly productive AI systems, the share of national income that continues to flow through wages, and whether democratic institutions can distribute the gains before economic power becomes much more concentrated.
How this tracker works
This is a living article, but it does not silently rewrite its predictions. New assessments are inserted above older ones, with the newest date first. If a judgement changes, the new section must say what changed and why. The older section remains intact so readers can see the actual history of the analysis.
TWIS monitors six underlying evidence groups: labour, ownership, distribution, capability, power and abundance. Evidence is recorded separately from interpretation. One company announcement or one dramatic headline is not enough to change a trajectory. A rating changes when evidence is repeated, independently supported or structurally important.
Weekly monitoring feeds the evidence ledger. A normal article update is prepared monthly. A major event can trigger an exceptional update between monthly editions. Publication remains review-gated: automation can prepare evidence and a pull request, but it cannot merge protected main.