The paper’s central argument is that the public debate has been too narrow. According to Jochen Büttner, coauthor from the Max Planck Institute of Geoanthropology, the AI debate revolves around safety, and the white paper agrees that red lines on models must be established and that liability for the companies publishing them must be enforced. But, he argues, this focus overlooks the fact that the most widespread harms of AI are fiscal and physical, and that AI companies are already harming people. The document therefore pairs the familiar safety agenda with two domains that regulators have treated far more lightly: the tax code and the environment. Its authors describe a technology sector that rewards the replacement of workers with software and that consumes land, water and fossil energy at a scale that local communities and national regulators are struggling to constrain.

The fiscal case rests on a structural mismatch. Tax systems were designed for an economy in which value comes from wages. In the United States, effective tax rates range from 25.5 to 33.5 percent on labour but sit at roughly 5 percent on equipment and software. Across the OECD, labour supplies about 57 percent of tax revenue while capital contributes about 19 percent. When a company replaces people with software, it cuts both its wage bill and its tax bill, and the public purse loses twice: once in revenue and again in the cost of supporting displaced workers. The exposure is large. The International Labour Organization estimates that one in four jobs worldwide, and 34 percent of jobs in high-income countries, is in an occupation exposed to generative AI. The white paper proposes ending tax rules that favour automation over employment, introducing a usage-based levy on compute, and returning part of the value of appropriated data to the public.

Jürgen Renn, another coauthor from the Max Planck Institute of Geoanthropology, frames the risk in blunt terms. The fiscal base, he says, is shifting from labour to capital faster than tax authorities can follow, and without reform AI will be paid for by cuts to the very services that cushion AI-generated unemployment and job insecurity. The paper’s taxation rule translates this diagnosis into legislative instruments. It calls for an end to tax treatment favouring automation over employment, a usage-based levy on compute or tokens collected on the destination principle, and a levy on data appropriation feeding a public fund. The proposed legal vehicles include member-state tax law with EU coordination, a UK Finance Bill modelled on the Digital Services Tax, and OECD frameworks.

The environmental chapter documents a sector moving in the opposite direction to its own earlier promises. Digital technology once promised dematerialisation: less paper, fewer journeys, smarter grids. The AI boom, the authors argue, goes the other way. On 6 October, a Finnish agency ordered Google’s local subsidiary to halt work at two data-centre sites after about 300 hectares of forest had been cleared without permission. In the United States, xAI runs about 1.5 gigawatts of gas turbines in Memphis, Meta has committed to ten gas-fired plants for its Louisiana campus, and 59 data-centre projects have announced some 90 gigawatts of onsite generation built on gas turbines and engines. In Spain, two thirds of data centres have water demand exceeding local supply in parts of the year. The International Energy Agency reports that data-centre electricity demand rose 17 percent in 2025, against 3 percent for electricity demand overall.

The United Kingdom, the paper notes, is accelerating consent for data centres in AI Growth Zones without binding conditions on efficiency, water or waste heat. The authors ask why big technology companies escape requirements that local European firms have always faced, and they answer with two regulatory proposals. The first is environmental limits on compute: data-centre consent and grid connection should be granted only with additional renewable supply, a power usage effectiveness of at most 1.2 and minimum waste-heat reuse, with consent withheld where impacts cannot be assessed. The proposed legal vehicles include Germany’s Energy Efficiency Act thresholds applied as permit conditions, the Irish grid-connection rule, planned EU minimum performance standards for 2027, and the UK National Policy Statement for data centres. The second is mandatory life-cycle disclosure: public energy and water figures per facility and per model, across training and inference, including upstream mining and electronic waste, achieved by amending Delegated Regulation (EU) 2024/1364 to move from confidential filing to publication and by extending the Energy Efficiency Directive’s Article 12.

Why Europe, and why now? The paper argues that the EU and UK together account for roughly a quarter of global AI demand, and that the so-called Brussels effect shows conditions attached to the common European market tend to become global corporate standards. The supply side matters too: every leading-edge AI chip is printed with lithography made only by ASML in the Netherlands, with optics provided by Carl Zeiss. Since 2 August, the European Commission has held enforcement powers over general-purpose AI models, and on 21 September it opened a consultation on minimum performance standards for data centres. The white paper’s conclusion is that the EU already has the legislative backbone and must now strengthen it in collaboration with the UK, rather than treat regulation as a purely American safety debate.

The thirteen rules are organised around one overarching instrument and twelve substantive measures spanning democracy, labour and equity, planetary health and safety. The centrepiece is a licence to operate for general-purpose AI in both the EU and the UK: no certificate, no market, with access conditional on demonstrated compliance and tightening on a published schedule, and exports of EUV lithography made conditional on downstream compliance. This would be implemented through Articles 51(2) and 97 of the AI Act by delegated act, the UK’s frontier AI bill, and Article 4 of the Dual-Use Regulation. Further rules address gatekeeping oversight, requiring fair ranking in AI answers, no self-preferencing and auditable retrieval logs, plus remuneration for authors whose works trained the models, using an extended Digital Markets Act Article 6(5), the UK’s digital markets conduct requirements and a review of the CDSM Directive. Competition rules would introduce systemic-importance designations keyed to compute share, merger approval conditional on open access, and procurement that rewards capability-extending AI.

Labour provisions call for human review of algorithmic pay and dismissal decisions, collective bargaining for data and moderation workers, and due diligence reaching contractors abroad, drawing on the Platform Work Directive and the Corporate Sustainability Due Diligence Directive. The safety rules retain the familiar red lines on containment escape, deception and shutdown resistance, pre-deployment evaluations with regulator-set thresholds in isolated environments, and liability resting with the model provider, implemented through AI Act Articles 5 and 55, statutory access for AI safety institutes and the Product Liability Directive. A final implementation rule would complete the standards layer and create a European AI Impact Observatory independent of the AI Office, using AI Act Articles 40, 41 and 68. The authors call on EU and UK leaders to add AI to the summit declaration, to agree a joint licence to operate, and to carry common standards for data centres and taxation into the UK’s G20 presidency in 2027.

The paper is published as a commentary rather than an empirical study, and its proposals are advocacy positions rather than evaluated policies. Its lead author, Felix Creutzig of the Bennett Institute at the University of Sussex, summarises the argument: Europe discusses AI as if the only question were safety in Silicon Valley, while the harm closer to home is a tax system that rewards replacing people with software and a digital sector that used to be green and is turning black, with forests cleared in Finland and gas turbines in Memphis. The 20 November summit, the authors argue, is the opportunity to put the rules on the table.

Subject of Research: A policy white paper proposing thirteen regulatory rules for artificial intelligence in the EU and UK, covering taxation, data-centre environmental limits, labour protections and model safety.

Article Title: Put AI on the table in Brussels: Thirteen rules for AI in Europe

Article References: Put AI on the table in Brussels: Thirteen rules for AI in Europe. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence regulation, EU policy, UK policy, data centres, taxation, environmental impact, generative AI, licence to operate, AI safety, labour market, Brussels effect, white paper