The Digital Monroe Doctrine: rebranding copyright infringement as a military asset

Abstract

This opinion argues that U.S. policy and industry efforts to normalise training AI on copyrighted works is the product of emerging security strategies. It shows how lobbying transforms creative labour into a geo economic asset, pressuring jurisdictions like the EU to accept lower standards or face otherwise regulatory retaliation. The piece compares the American path against EU safeguards of Directive (EU) 2019/790 of the European Parliament and the Council of 17 April 2019 on copyright and related rights in the Digital Single Market (hereinafter CDSM) and the AI Act’s transparency and compliance obligations, while noting the practical weaknesses. It recommends sovereign reciprocity through collective remuneration to compensate rights holders, a mandatory interoperability standard for machine-readable rights reservations and independent pre-deployment audits as a condition for EU market access. These measures would protect creators, close enforcement gaps and use Europe’s market power to align innovation incentives with fundamental rights and European values.

Keywords: copyright, fair use, AI Act, security, innovation

Introduction

The United States has a long history of declaring strategic zones of influence and warning rivals to stay out. In 1823, President James Monroe told European powers that the Western Hemisphere was closed to colonial interference. Two centuries later, Washington has done the same in the digital domain. Geographical boundaries have given way to intellectual property concerns and America’s leadership in AI has become the new hemisphere to defend (Berger, 2025).

Copyright reflects choices about who creates value, who owns it and who profits from it. Today, the Trump administration is reframing those choices as questions of national survival. What was once copyright infringement is now a military asset in the race to contain China. What was once a rightsholder’s legitimate interest is now an obstacle to Western technological dominance. What was once the EU’s human-centric AI framework is now perceived as taxation on American success. Europe must not mistake this pressure for a mere legalistic argument. It is a geopolitical strategy and it deserves a geopolitical response.  

Securitisation as a strategy

In March 2025, OpenAI and Google submitted policy proposals to the White House requesting that AI training on copyrighted material be classified as fair use (Wiggers, 2025). When DeepSeek-R1 launched in January 2025, it matched the performance of GPT-4 class models at a reported training cost of $6 million, against OpenAI's estimated $100 million for the equivalent (OpenAI, 2025). As a result, Silicon Valley responded with lobbying. OpenAI subsequently accused DeepSeek of distilling US models for their own model training, characterising it as “free-riding” on American capabilities (Seetharaman and Arámburo, 2026). The conclusion was clear: if China breaks the rules, American firms should be permitted to bypass different ones.

By summer 2025, Andreessen Horowitz (a16z) and OpenAI president Greg Brockman had committed $100 million to “Leading the Future”, a super Political Action Committee (PAC) explicitly lobbying against strict AI regulation (Lau, Tolentino and Lennet, 2025). Likewise, Microsoft, Amazon, Alphabet and Meta, led by the trade association Incompass, jointly lobbied for an allegedly ten-year federal ban on state AI legislation (Gallagher, 2025). The Trump administration’s National AI Policy Framework responded accordingly. Federal law seeks to proactively align with the tech industry’s demands, in an attempt to preempt state AI laws such as California’s Senate Bill 53 or New York’s Senate Bill 7263 (Kraus, 2026). As a result, AI training on copyrighted content will be treated as lawful, broadening the fair use doctrine as a strategic asset. Nonetheless, the US Copyright Office’s May 2025 report established that certain uses of copyrighted material for AI training cannot be defended unless the transformative nature can be proven (Ginsburg, 2025, pp 521-522). Against this, the Trump administration seeks to pressure for a new Register of Copyrights to adopt a different position.  

What Fair Use implies

Under USA copyright law, the fair use doctrine does not enumerate an exhaustive list of prohibited activities but rather establishes a case-by-case balancing test. There are four key factors that have to be considered: the purpose and character of the use, the nature of the copyrighted work, the amount used and the effect on the potential market for the original work (Buick, 2025, p. 187). Transformation carries particular weight and use is more likely to qualify as fair if it adds something new, with a different purpose or character.

Creative industries have already fought this battle. Hollywood writers strive to prevent AI- generated material from undermining their credits or separating rights (Coyle, 2023). Moreover, over 200 artists, including Billie Eilish and Kacey Musgraves, publicly accused AI developers of using their music without their consent or authorisation to train models  “directly aimed at replacing the work of human artists” (Roeloffs, 2024). These are not isolated claims, but rather a growing reality that has spread across different sectors.  

It wasn’t until the 2025 copyright case, Thomson Reuters v. Ross Intelligence, that a court rejected the fair use defence in the context of training an AI system. The tribunal found that Ross’s use of Thomson Reuter’s headnotes was not transformative, because it aimed to build a competing legal research product rather than add new expression or meaning. Judge Bibas warned explicitly that allowing unrestricted scraping of copyrighted content would undermine intellectual property law. Although the decision was case-specific and it did not address generative AI, its reasoning is relevant to the industry’s core argument (Daniel, 2025). It is important to note that Ross filed an interlocutory appeal in April 2025, which means the case remains pending and its precedential value has to be determined.  

European sovereignty and data leaking

The EU operates within a distinct regulatory framework. Article 4 of the CDSM Directive permits commercial text and data mining but grants rightsholders an opt-out through machine-readable reservation of rights. Articles 53(1)(c) and (d) of the AI Act require general-purpose AI model providers to maintain a copyright compliance policy and publish a detailed training data summary (Ziaja, 2024). In theory, transparency enables enforcement; in practice, it facilitates evasion. No standardised protocol for machine-readable opt-out currently exists, leaving individual authors without a viable mechanism to verify compliance. As Buick (2025) concludes, transparency is a necessary but insufficient condition, as it is designed to support an opt-out mechanism that remains structurally broken.

The territorial gap is more acute since copyright is geographically delimited. If AI training occurs outside the EU, no infringement of EU Member State copyright is triggered. Recital 106 of the AI Act asserts that the opt-out obligation applies to any provider placing a model on the EU market regardless of training location (Quintais, 2025, p. 13). Through the extraterritorial application known as the Brussels Effect, the European Union imposes a strict “market entry requirement” as a condition to enter its regulatory space (Abbamonte, 2024, p. 485). Building on this logic, authors such as Rosati (2024) argue that text and data mining are an inherent component of the deployed model and thus connected to the EU market (p. 615). Critics claim that this practice would incentivise US companies to relocate their training activities to countries with less stringent regulatory norms. As a result, instead of setting global standards, the AI Act would produce a counter-productive effect.

The weaponisation of innovation

Trump has characterised EU AI regulation as an obstacle to American success and an act of economic hostility. In August 2025, he threatened retaliatory tariffs against countries imposing digital regulation on US companies (Martin, 2025). In November 2025, the European Commission presented the Digital Omnibus package which streamlines certain  obligations of the AI Act and GDPR, although other provisions such as bias detection are claimed to have expanded (Niestadt, 2026, p. 3). These modifications further widen the regulatory asymmetries that function as industrial policy in favour of American companies.

It seems bizarre that a creator would consent to their work being embedded into infrastructure used for geopolitical motives they never consented to. If the Trump administration successfully rebrands this extraction as “patriotism”, then innovation becomes a state- sanctioned act that strips the creative class of the ownership of their work. The exodus of talent becomes an unavoidable consequence. European AI innovation and researchers stand at a crossroads: build under restrictive compliance regimes with higher data acquisition costs or relocate to the US, where capital is abundant and regulation is permissive.

Policy statement: from defensive legalism to sovereign reciprocity

Europe must recognise that the US is not merely stating a legal argument about fair use doctrine, but it is establishing who will hold technological hegemony in the upcoming decades. Within this framework, copyright is not perceived as a legal obligation but rather a strategic liability that must be eliminated. Europe’s response must operate at the same level of ambition. Recommendations must reconcile rightsholders’ economic interests and the efforts to boost technological innovation.

First recommendation

The most problematic is the opt-out clause from Article 4(3) since the mechanism places the burden on the individual author. Such is expected to identify use, reserve rights and enforce compliance against some of the most powerful existing companies. Imposing a levy, requiring AI developers to contribute to a collectively managed fund compensating rightsholders, would align incentives without requiring individual enforcement (Buick, 2025).

Second recommendation

There is no agreed technical standard for machine-readable rights reservation, which renders Article 4 CDSM structurally inoperable in practice. The European Commission must require an interoperability standard binding on all general-purpose AI model providers that access the EU market. This directly complements the documentation requirements already imposed by the AI Act: rather than merely advocating for transparency about the extraction of data, a mandatory standard would simultaneously prevent unauthorised and unconsented extraction at the technical level. Transparency without prevention is documentation of harm, not its remedy.

Third recommendation

Access to the EU market must rest on verifiable training data compliance, not self- declaration. Independent audits must function as a sine qua non instrument: providers seeking EU market access should be required to submit externally verified training data summaries  before deployment, not after enforcement action. This is the operational extension of Recital 106 of the AI Act’s extraterritorial claim. Market power is Europe’s most effective tool and therefore must be used at the point of entry, not retrospectively.

Conclusion

A Digital Monroe Doctrine is taking shape in Washington and the intentions are simple: American technology companies will train on whatever data they require. Any government that enforces its copyright laws against that practice is engaging in economic aggression.

The risk for Europe is not merely legal. It is existential. If the Brussels Effect is neutralised by regulatory arbitrage and political pressure, European startups face compliance costs that Americans do not. And if the talent that could build a competitive European AI sector continues to migrate West, Europe will not merely lose a trade argument. It will lose the capacity to build its technological sovereignty. Washington has declared that copyright infringement, rebranded as patriotism, is a national security asset. In the midst of this “Digital Cold War”, Europe must be equally clear in response. The rights of creators are not negotiable and the systematic use of creative work without the author’s consent is a political choice, not an expression of technological development.  

Bibliography

Abbamonte, G. B. (2024). The application of the copyright TDM exceptions and transparency requirements in the AI Act to the training of generative AI. European Intellectual Property Review, 46(7), 479–487.

Berger, V. (2025, March 17). The Ai Copyright Battle: Why OpenAI and Google are pushing for Fair Use. Forbes. https://www.forbes.com/sites/virginieberger/2025/03/15/the-ai- copyright-battle-why-openai-and-google-are-pushing-for-fair-use/

Buick, A. (2025, December 12). Copyright and AI training data—transparency to the rescue? Journal of Intellectual Property Law & Practice 20(3), 182-192. https://doi.org/10.1093/jiplp/jpae102

Coyle, J. (2023, September 27). In Hollywood writers’ battle against AI, humans win (for now). AP. https://apnews.com/article/hollywood-ai-strike-wga-artificial-intelligence- 39ab72582c3a15f77510c9c30a45ffc8

Ginsburg, J., C (2025, August 4). AI inputs, fair use and the US Copyright Office Report. Journal of Intellectual Property Law & Practice, 20(8). 10.1093/jiplp/jpaf046

Gallagher, W. (2025, June 18). Meta, Amazon, and more want 10-year ban on states regulating AI. Appleinsider. https://appleinsider.com/articles/25/06/18/meta-amazon-and- more-want-10-year-ban-on-states-regulating-ai

Kraus, J. (2026, April 10). White House AI Framework Proposed Industry-Friendly Legislation. Lawfare. https://www.lawfaremedia.org/article/white-house-ai-framework- proposes-industry-friendly-legislation

Daniel, L. (2025, February 14). U.S. court ruling puts AI training practices under the microscope. Forbes. https://www.forbes.com/sites/larsdaniel/2025/02/14/us-court-ruling- puts-ai-training-practices-under-the-microscope/

Lau, R., Tolentino J.J., Lennet B. (2025, December 3). November 2025 US Tech Policy Roundup. Tech Policy. https://www.techpolicy.press/november-2025-us-tech-policy- roundup/

Martin, N. (2025, January 9). Digital taxes put US-EU trade talks under pressure. DW. https://www.dw.com/en/digital-taxes-put-us-eu-trade-talks-under-pressure/a- 73798948

Niestadt, M. (2026, June). Digital Omnibus on AI. European Parliament. https://www.europarl.europa.eu/RegData/etudes/BRIE/2026/782651/EPRS_BRI(202 6)782651_EN.pdf  

OpenAi. (2025, March 13). OpenAi’s proposal for the U.S AI Action Plan. OpenAi Global Affairs. https://openai.com/global-affairs/openai-proposals-for-the-us-ai-action-plan/.

Quintais, P., J. (2025, April). Generative AI, copyright and the AI Act. Computer&Law Security Review, 10. https://doi.org/10.1016/j.clsr.2025.106107.

Roeloffs, M. W. (2024, April). Artists Slam Developers For Using Music Without Permission In Letter Signed By Kacey Musgraves, Billie Eilish And More. Forbes. https://www.forbes.com/sites/maryroeloffs/2024/04/02/artists-slam-ai-developers-for-using- music-without-permission-in-letter-signed-by-kacey-musgraves-billie-eilish-and-more/  

Rosati, E. (2024, 31 October). Infringing AI: liability for AI-generated outputs under International, EU, and UK Copyright Law. European Journal of Risk Regulation 16(2), 603- 627. https://doi.org/10.1017/err.2024.72

Seetharaman, D., Arámburo F. (2026, February 13). OpenAI says China's DeepSeek trained its AI by distilling US models, memo shows. Reuters. https://www.reuters.com/world/china/openai-accuses-deepseek-distilling-us-models-gain- advantage-bloomberg-news-2026-02-12/  

Wiggers, K. (2025) OP. OpenAI calls for US government to codify “fair use” for AI training. Techcrunch. https://techcrunch.com/2025/03/13/openai-calls-for-u-s-government-to-codify- fair-use-for-ai-training/

Ziaja, G.,M. (2024). The text and data mining opt-out in Article 4(3) CDSDM: Adequate veto right for rightholders or suffocating blanket for European artificial intelligence innovation? Journal of Intellectual Property Law, 19(5), 453-459. https://doi.org/10.1093/jiplp/jpae025  

Moderate use of AI models: yes/no – language editing and source consultation