The only pillow that catches you when AI takes your job.
Ordering means signing a petition to your government. With state products there is no other way. · skip to the data ↓
Quality stitching (the benefit replacement rate) can absorb up to two thirds of the drop in consumption after losing a job.
Gruber, AER 1997 · in the optimal configurationThe layer of active labour market policies takes 2-3 years after impact to bear full load, and it works best precisely in a recession.
Card, Kluve & Weber, meta-analysis of 200+ programmesA cushion takes decades to sew: taxes, laws, retraining. Once the impact hits, you cannot reorder.
Autor, Dorn & Hanson: adjustment takes over a decadeWe do not ship to individuals. Your cushion's thickness is set by the system of the country you work in.
In the long run the system decides, not the individualNo cushion in this catalog is built for a shock of that size. Not even the Danish one. This map does not measure who is done. It measures who has something to build on and who is sewing from scratch. A cushion takes a decade to sew. That may be exactly how much time is left.
Thickness matches your country's cushion index: adaptive capacity + social protection, live Eurostat data.
Most of them do not know it. The star rating equals the cushion thickness of their country.
Your edition was assigned by the state. The only way to change it is to change the system, or the country.
Mandatory disclosure: The Anti-AI Pillow is not a product. It is the state of your country's social system, measured from Eurostat (2023-24), OECD and ICTWSS data. None of it is for sale. All of it can be ordered, below.
No trick: an "order" is your signature under a petition to your country's government. Its full text is below. Read what you are signing.
Target specification = the Danish model (the thickest in the catalog). Delivery depends on your government: lead times run in years, so order today.
No payment. The price is already paid, through your taxes. (Mockup: signatures are not submitted yet. In production we will connect a partner petition platform, including the legal requirements per country.)
order no. ·
Status: waiting for the government. Once enough signatures come in, we deliver them. Tell others about the shop: a cushion gets sewn faster when the whole country is ordering.
A region's safety ≠ low exposure. We read safety as the ratio of two forces: pressure, meaning how much of the local economy stands on occupations that today's AI can perform, and the cushion, meaning people's adaptive capacity and the state's systemic social protection. The primary reading is the quadrants, not a single number. The most dangerous place on the map is the one where nobody is worried.
In July 2026, 16 Nobel laureates and over 200 economists and AI builders signed a call warning that the transformation may be "bigger than the industrial revolution, compressed into a shorter time", and that action must come now. Our map does not contradict that; it is one of its instruments: the quadrants do not measure who has enough, but who has something to build on. A strong cushion in a European comparison is a cushion built for the shocks Europe knows, not for the shock the letter warns about. That is why our petition and the Ten Demands apply to all 30 countries, including the Head Start quadrant: the difference between countries is not "done vs. not done" but "sewing vs. not sewing".
Quadrants are relative positions, not absolute safety. We split at medians, so half of Europe is "above the line" by definition: the map says who is more prepared than their neighbours, not who is prepared. And we measure the cushion with instruments calibrated to last century's shocks: the verified evidence says that even in the best systems, adjustment to a large shock takes over a decade (Autor, Dorn & Hanson), transfers do not compensate income losses, and benefits at their optimum absorb about two thirds of the consumption drop, not the fall itself. That is why the lime quadrant is called Head Start, not safety. On top of that, 79 million European workers (38%) live in a country whose cushion is weak even by today's standards.
A. Exposure (pressure). The occupational mix of employment (Eurostat EU-LFS, ISCO-08) weighted by the AIOE exposure index. Planned second layer: observed exposure from the Anthropic Economic Index, reading the gap between theoretical and observed exposure as "runway".
B. Adaptive capacity. Tertiary education, lifelong learning, unemployment (inverted), NEET (inverted), R&D expenditure. All live from Eurostat.
C. Social protection. Social protection expenditure (ESSPROS, live), benefit replacement rate (OECD TaxBEN), collective bargaining coverage (OECD/ICTWSS), active labour market policies (LMP). National level: the net is national, which is a feature, not a bug.
Composition. Cushion = the geometric mean of adaptation (B) and protection (C): √(B·C). Adaptation and protection are complements, not substitutes: a pillow with huge filling on one side and a hole on the other is not half a good pillow, and the geometric mean punishes the weak link. That is why we always show B and C separately as well. Normalization: winsorized min-max (5th-95th percentile), rescaled to 0.05-1. Weights = evidence-informed indicator selection + explicit expert weighting; they remain publicly adjustable (panel below).
Quadrants. Split at the standard European median; around the dividers we draw an uncertainty band (between the 15th and 16th country). Scientifically these are relative categories HH / HL / LL / LH; the names "Head Start", "Critical Zone", "False Safety" and "Stable Base" are editorial labels: the campaign's interpretive layer, not methodological judgements. A relative position does not mean absolute preparedness.
Uncertainty as part of the result. We ran every country through 5,000 model specifications (Monte Carlo: weights, B:C ratio 40-60, arithmetic/geometric/harmonic aggregation, two normalizations, ± noise on exposure weights, ±15% on provisional series, leaving out indicators and countries). For each country we report the share of specifications in its quadrant: Czechia is in LL in 100% of specifications, Cyprus in HH in only 46% (so read it as borderline). The exposure × cushion relationship also survives a control for GDP per capita (partial r = 0.61), so the index does not merely measure the gradient of economic development.
Relation to existing work. The two-dimensional exposure × preparedness architecture builds on the IMF framework (Cazzaniga et al. 2024, Gen-AI and the Future of Work). Our narrowing: we build the second axis specifically on the labour market's ability to absorb a work shock, and we separate the workforce's adaptive capacity (B) from ex-post social protection (C), with fully transparent and user-testable assumptions.
We are not selling a prediction of professions dying out. Exposure ≠ replacement: we talk about pressure and a cushion, not about "lost jobs". Corporate layoff attributions are not an input to the index; at most a narrative layer with an "AI washing" caveat.
1) Exposure index: AIOE vs. Anthropic AEI vs. both. 2) UK and the Balkans in v1? 3) One public number, or quadrants only? 4) Final project name.
5 of 8 index indicators come from an official source (Eurostat, snapshot 7/2026, ref. 2023-24); 3 are provisional (flagged in the detail). NEET and R&D are context layers outside the index. Quadrants = relative positions against the European median.
Equal weights are the v0 default and a deliberately public decision. Changing the weights recomputes the cushion, the medians and the quadrants. Pressure (pillar A) is a single indicator for now, so it carries no weight slider.
What exactly we ask of governments: our answer to the "We Must Act Now" call (16 Nobel laureates and 200+ economists, 7/2026). Every point stands on verified evidence: no castles in the air, just budget lines. It applies to all 30 countries, including those with a head start.
A public annual stress test of the social system against an AI shock, on data, not impressions. What is not measured does not get sewn. The methodology of this index is open source.
Raise spending on retraining, counselling and back-to-work support. They work best precisely in a crisis, yet Czechia spends 0.18% of GDP where Denmark spends 1.4%. Card, Kluve & Weber, meta-analysis of 200+ programmes.
Individual learning accounts portable between employers, so that people in exposed occupations train before they fall. Today they train 3× less than everyone else. OECD, Nedelkoska & Quintini.
An unemployment benefit replacement rate high enough that losing a job does not mean an immediate drop in living standards. Well-set benefits absorb up to two thirds of the shock. Gruber, AER.
The self-employed, platform and agency workers need the same net. AI does not select by contract type.
Everyone under 30 who has been out of work and school for 6 months gets a concrete offer of a job, an internship or education. NEET is the most expensive line item of the future.
Strengthen worker consultation when AI is deployed: where people are talked with, AI adoption demonstrably turns out better for the firm as well. OECD Employment Outlook 2023.
Large firms report the impact of AI deployment on jobs. Without data on the pressure, no cushion can be tailored.
Automation gains must flow into financing the cushion. The tax mix cannot rest only on taxing the very work that is disappearing.
Labour market adjustment to a major shock takes over ten years. Approve the cushion's budget and plan on a ten-year horizon. Autor, Dorn & Hanson.