The European Parliament just signaled a seismic shift in how AI companies must pay for training data. On March 10, 2026, lawmakers adopted a resolution proposing a flat-rate copyright fee of 5 to 7 percent of global turnover for any AI system offered in Europe that uses creative works for training. This could fundamentally reshape how Stability AI, Midjourney, and other image generation platforms operate worldwide. Why Is Europe Targeting AI Training Data? The resolution reflects a growing frustration among European policymakers and creative industries that generative AI systems like Stable Diffusion have been trained on billions of copyrighted images, articles, music, and videos without explicit permission or compensation. Traditional copyright law, designed for an era of printing presses and individual works, doesn't adequately address how modern AI systems process massive datasets containing billions of digital copies. The core tension is straightforward: AI companies need high-quality training data to build powerful models, but photographers, journalists, musicians, and filmmakers argue their creative work is being used without authorization or fair compensation. The European Parliament's Committee on Legal Affairs concluded that existing copyright frameworks are insufficient and that additional legal mechanisms are needed to balance innovation with creator protection. What Would This Tax Actually Look Like? The proposal is remarkably specific. The Committee Report suggests establishing an "immediate, simple, flat-rate copyright fee" of 5 to 7 percent of global turnover to compensate European creators. This means if Stability AI generates $100 million in annual revenue from its image generation services, the company could owe $5 to $7 million annually to a fund benefiting European creative industries. What makes this proposal particularly significant is its potential retroactive application and global reach. The resolution suggests this fee could apply not just to revenue earned in Europe, but to the entire global turnover of any AI system offered to European users. For companies like Midjourney, which operates a subscription model ranging from $10 to $120 monthly, this could translate to substantial financial obligations. How Are Courts Already Grappling With This Issue? The legal uncertainty surrounding AI training is already visible in conflicting court decisions across Europe. In November 2025, the Regional Court of Munich ruled in a case brought by GEMA, a German collecting society, against OpenAI that certain forms of AI training involving protected works could constitute copyright infringement and that the EU's text-and-data-mining exception did not apply. However, that same month, the High Court in London dismissed Getty Images' copyright infringement claim against Stability AI, concluding that the Stable Diffusion model could not be considered an infringing copy under applicable legal standards. These diverging outcomes illustrate that courts across European jurisdictions are still grappling with how traditional copyright concepts should apply to generative AI technologies. The European Parliament's resolution appears designed to settle this uncertainty through legislation rather than waiting for courts to establish precedent. What Practical Changes Could This Trigger? If implemented into binding legislation, the proposal would require significant operational changes for AI companies. The resolution suggests creating standardized opt-out mechanisms allowing rights holders to signal that their content should not be used for AI training, with signals being machine-readable for automated compliance. This would likely require: - Data Licensing Infrastructure: AI companies would need to establish systems for identifying, tracking, and licensing copyrighted content used in training datasets, similar to how music streaming services manage royalty payments. - Registry Systems: The proposal mentions creating a potential European Union Intellectual Property Office (EUIPO) registry where creators can register opt-out preferences, requiring AI companies to check this registry before using content. - Financial Restructuring: Companies would need to either absorb the 5-7 percent revenue tax, pass costs to users through higher subscription prices, or reduce spending on model development and infrastructure. What's the Tension Between Innovation and Protection? The European Parliament's approach reflects a genuine policy dilemma. On one hand, the European Union seeks to strengthen its competitiveness in artificial intelligence and reduce the technological gap with the United States and China. Achieving this requires powerful AI models and access to large quantities of high-quality training data. On the other hand, media companies, publishers, musicians, filmmakers, and digital platforms produce much of the content that trains generative AI systems, and many argue they deserve compensation for this value creation. This explains why the resolution attempts to balance both objectives: regulating AI training while protecting the economic sustainability of Europe's creative industries. However, the proposal's global reach could create friction with non-EU legal regimes governing AI training practices, potentially leading to tensions between European and American approaches to AI regulation. How Might This Affect Specific AI Image Generators? Different image generation platforms would face different impacts. Midjourney v6, which operates as a closed-source service with no public model weights or API access, generates revenue through subscription tiers ranging from $10 monthly for basic access to $120 monthly for the Mega tier. A 5-7 percent global revenue tax would directly reduce profitability unless passed to users. Stable Diffusion, which offers both open-source and commercial versions, might face compliance challenges given its training on LAION-style datasets containing billions of internet images. The proposal's emphasis on retroactive compensation suggests that companies could face claims for past training activities, not just future ones. This creates significant financial uncertainty for established AI companies that have already invested billions in model development based on existing copyright interpretations. What Happens Next? It's important to note that the European Parliament's resolution itself is not legally binding. However, it signals the political direction of EU institutions and likely foreshadows future legislative proposals from the European Commission. The resolution's adoption suggests that binding legislation implementing some form of copyright compensation for AI training could emerge within the next 12 to 24 months. For AI companies operating globally, this represents a critical inflection point. The outcome in Europe could influence regulatory approaches in other jurisdictions, potentially creating a domino effect where other regions adopt similar frameworks. Alternatively, if the proposal faces industry pushback or proves difficult to implement, it could signal that copyright reform for AI training remains contentious and uncertain. The fundamental question remains unresolved: Should AI companies pay for the creative works they use to train powerful models, and if so, how much? Europe appears ready to answer that question through legislation, even if the answer creates friction with companies and legal regimes outside its borders.
