Researchers propose an AI compact for Europe built around tax, power and market access

A group of 17 researchers wants European AI policy to reach beyond model testing and into the economics of deployment. A thirteen-rule framework, promoted by the Max Planck Institute of Geoanthropology on 27 October, connects market access with taxation, resource use and public accountability. The proposals would change the conditions under which AI is built and sold; they are not rules governments have adopted.[1]

The chronology matters. The white paper is dated 12 October, and the Bennett Institute announced it on 22 October. The latest release is a renewed policy intervention, not the enactment of a new European law. Its four substantive themes are democracy and agency, labour and equity, planetary health, and safety.[3][2]

Market access as the enforcement mechanism

The framework’s organising idea is a licence to operate: developers would have to demonstrate compliance before accessing European markets. Bennett’s explanation argues that the EU and UK can exert influence through their purchasing power and roles in the AI supply chain. The authors favour systems that extend human capabilities rather than treating the replacement of labour as the default measure of progress.[2]

That is a coherent enforcement theory, but its effectiveness would depend on implementation. A regulator needs observable conditions, credible inspections and consequences for non-compliance. A certificate that verifies little becomes paperwork; an approval process that smaller entrants cannot navigate may protect incumbents. These are design tensions, not evidence that the proposal has already succeeded or failed.

Paying for the costs outside the model

The fiscal proposals include a usage-based compute or token levy and a charge for data appropriation that could support a public fund. The paper also proposes addressing local energy, noise and water costs. Its premise is that the public consequences of deployment should appear in the incentives facing the organisations deploying it.[3]

Choosing the tax base would be consequential. A levy on a billable unit of AI use is administratively different from a charge on a facility’s water consumption. One raises revenue from a service; the other targets a local physical burden. A single measure should not be assumed to do both jobs well merely because both involve AI.

The Max Planck announcement places that argument alongside the balance between labour and automation, and the distribution of gains from AI. Its authors want fiscal policy to influence the terms of adoption, not just collect revenue after the commercial decisions have been made. That is an argument about who benefits, as well as about how safely a model behaves.[1]

Efficiency and expansion are different questions

For new data centres, the paper proposes strict efficiency conditions, waste-heat reuse and renewable additionality: new demand would have to be matched by new renewable capacity. It also advocates withholding permission where environmental effects cannot be assessed with reasonable confidence. These are proposed permit conditions, not a description of a uniform rule now operating throughout Europe.[3]

The distinction between efficiency and total scale is easy to illustrate. A hypothetical facility can cut the energy needed per unit of computation by one-fifth yet double its output. Its total energy use would then rise by 60%: 0.8 multiplied by 2 equals 1.6. This is an illustrative calculation, not a forecast for a particular operator.

Disclosure, price signals and permission to expand therefore solve different problems. Disclosure makes a burden visible; a levy changes its financial cost; a permit condition can limit it outright. Combining them may be sensible, but overlapping requirements need consistent definitions and an accountable authority to resolve conflicts.

The framework’s value is to make those choices explicit. Instead of treating AI governance as a boundary around the model alone, it asks how market access, public revenue and physical infrastructure should work together. The difficult political task would be turning that shared ambition into requirements that can be measured, enforced and challenged—not merely announced.

Report an error or suggest a correction