The Need for Standardization in TRIPS Exhaustion Policy for Use of Copyright-Protected Data for TDM

Yue Chu December, 2024

Share with:

Under TRIPS, the doctrine of exhaustion is applied to patents, pharmaceuticals, and physical goods, but its application to digital material is a legal gray area.[1] The doctrine establishes that ownership of tangible goods is transferred with a single, definitive transaction, and naturally, the new owner has full control while the previous owners lose theirs.[2] Digital data, on the other hand, can be replicated with no loss or reduction of the original copy, so the data can be distributed without authorization.[3] Access to digital data is shared rather than definitively lost or gained. Traditionally, the doctrine of exhaustion states that when a copyrighted work is sold, the original owner loses the right to control the distribution of that copy; however, this works less clearly for digital data.[4] TRIPS Article 6 discusses the principle of exhaustion but does not reference data exhaustion and allows member states to establish their own scope and context for exhaustion by implementing their own copyright laws to address the issue.[5] With the rise of AI, a lack of standardization in international data exhaustion policy affects cross-border cooperation and hinders technological progress.[6] AI technology relies on text and data mining (TDM), and TRIPS offers no guidance for how rights to this data are exhausted.[7] This leaves member states free to apply their own differing copyright protection laws to the replication and distribution of data for TDM, which leads to irreconcilable differences in cross-border transactions.[8]

Under the liberal copyright laws in the US, when data is used properly for TDM, original rights to data are functionally exhausted the same way rights to tangible goods are.[9] If data is lawfully acquired, regardless of the copyright holder’s intent, the new owner may defend TDM as fair use as long as they demonstrate their usage satisfies the four fair-use factors: transformative, nature of copyrighted work, amount/substantiality of the portion copied, and effect on the market for the original work.[10] No permission or licensing from the original rightsholder is required, functionally exhausting the rightsholder’s control over the data’s distribution. For example, in Author’s Guild v. Google, the Second Circuit concluded that Google’s unauthorized digitizing of copyright protected works and limited display of those works counted as non-infringing fair uses, and Google’s commercial profit from these actions is irrelevant.[11] This decision paints unauthorized copying of data for transformative TDM as a non-infringing fair use. On the other hand, under the EU’s Digital Single Market (DSM) directive, TDM is restricted to non-commercial, scientific research in Article 3.[12] Though Article 4 expands TDM usage to commercial uses as well, original rightsholders of the data may still opt out of letting their data be used for commercial purposes.[13] The EU generally requires licenses for TDM activities, which gives the original rightsholder a chance to govern the terms under which the data can be used. This means the rights of the original owners of the data are not exhausted when the data is distributed to a new user, as a result, original data owners retain control over how their data is used for TDM purposes.[14] In the German case Kneschke v. LAION, the Hamburg regional court ruled that LAION’s use of Kneschke’s photographs for AI training counted as a scientific use and was thus allowed without licensing under Section 60d of the German Copyright Act (UrhG), which implemented DSM Article 3.[15] However, the court noted that creators may assert opt-out options against non-scientific commercial use under Section 44b of UrhG, which the court ultimately decided was not the case here.[16] In US, commercial use by the new user does not expand the scope of protection for the original user, unless that commercial use encroaches negatively on the original user’s market per factor four of the fair use doctrine.[17]

The ambiguity in the doctrine of exhaustion under TRIPS allows countries with more liberal copyright laws, like the US, to essentially treat digital data as “exhaustible” as physical goods, while countries in the EU with more stringent copyright protection do not fully exhaust the rights of the original owner of the data after the data is transferred.[18] Disputes are emerging already. In Getty Images v. Stability AI, Getty Images has sued Stability AI in both the USA and the UK for training its AI model on copyrighted images without permission.[19] It is unclear whether Stability AI had conducted its training in the US or the UK.[20] Stability AI alleges that its training was done on US servers, which would be beneficial for its defense, since the US may allow this under fair use.[21] Getty Images maintains that some of the data was transferred and processed within the UK, which would violate the copyright protection afforded by the DSM.[22] If this dispute is treated inconsistently between the jurisdictions, litigation would be extremely problematic. Having Stability AI’s training practices be legally permissible in the US but prohibited in the EU would undermine the harmonization goal of TRIPS and set a harmful precedent that encourages forum shopping, whereby developers cherry-pick countries with more lenient copyright laws to train AI models.[23] Furthermore, any ruling invites the parties to clash over Article 13 of TRIPS, which sets the exceptions to copyright protection: exceptions must be tailored to certain special cases, have no conflict with normal exploitation, and contain no unreasonable prejudice to legitimate interests.[24] Stability AI could argue that the EU’s allowance of opt outs under DSM Article 4 prevents normal exploitation, which includes fair use in the US. Getty Images could argue that the US’s flexible, case-by-case determination of fair use is too amorphous a policy to be considered “specifically tailored” under Article 13.[25]

Under US law, Getty Images’ rights to the data are exhausted if Stability AI establishes fair use, while under the EU’s DSM Article 4, Getty Images’ rights are not.[26] A TRIPS amendment to Article 6 should clarify the doctrine of data exhaustion and explain the difference between data and tangible exhaustion. Article 6 currently allows member states too much freedom to define “exhaustion.” If different member states are allowed to define “exhaustion” in accordance with their varying policies on copyright protection, data exhaustion could be treated differently by different states, leading to no consensus in cross-border disputes. Article 6 should include an amendment that defines data exhaustion and clarifies whether it is treated the same as exhaustion of physical goods, or if there are exceptions that protect the original rightsholder’s control over the data.

  1. Pascale Chapdelaine, Copyright User Rights: Contracts and the Erosion of Property 109 (2017).↑

  2. Id. ↑

  3. See Aaron Perzanowski & Jason Schultz, Digital Exhaustion, in 58 UCLA L. Rev. 889-946 (2010), available at SSRN: https://ssrn.com/abstract=1669562.↑

  4. PA Legal, What is the Doctrine of Exhaustion In Copyright Law?, https://thepalaw.com/copyright/what-is-the-doctrine-of-exhaustion-in-copyright-law/ (last visited Oct. 25, 2024).↑

  5. World Trade Organization, Agreement on Trade-Related Aspects of Intellectual Property Rights art. 6, Apr. 15, 1994, https://www.wto.org/english/docs_e/legal_e/27-trips_03_e.htm.↑

  6. See Jana Kazaz, Ethical Use of AI: Navigating Copyright Challenges, GLOBSEC, https://www.globsec.org/what-we-do/publications/ethical-use-ai-navigating-copyright-challenges (last visited Oct. 25, 2024).↑

  7. See World Trade Organization supra note 5.↑

  8. Peter Yu, TRIPs and its Discontents, 10 Marq. Int’l L. Rev. 370, 387 (2006) (discussing the international inconsistency in the exhaustion doctrine).↑

  9. Antti Vaihia, Copyright Laws and Machine Learning, Valohai Blog (Mar. 27, 2023), https://valohai.com/blog/copyright-laws-and-machine-learning/.↑

  10. See Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015).↑

  11. Id.↑

  12. Directive 2019/790, of the European Parliament and of the Council of 17 April 2019 on Copyright and Related Rights in the Digital Single Market, art. 3, 2019 O.J. (L 130) 92.↑

  13. Directive 2019/790, of the European Parliament and of the Council of 17 April 2019 on Copyright and Related Rights in the Digital Single Market, art. 4, 2019 O.J. (L 130) 92.↑

  14. See id.↑

  15. Ronak Kalhor-Witzel, A Landmark Copyright Case with Implications for AI and Text and Data Mining: Kneschke v. LAION, Trademark Lawyer Magazine (Sept. 27, 2024).↑

  16. Id.↑

  17. See Google, Inc. supra note 10.↑

  18. See P. Bernt Hugenholtz, The New Copyright Directive: Text and Data Mining (Articles 3 and 4), Kluwer Copyright Blog, (July 24, 2019), https://copyrightblog.kluweriplaw.com/2019/07/24/the-new-copyright-directive-text-and-data-mining-articles-3-and-4/.↑

  19. Complaint, Getty Images (US), Inc. v. Stability AI, LTD. No. 1:23-cv-00135, at 3 (D. Del. Feb. 6, 2023); Amended Complaint, Getty Images (US), Inc. v. Stability AI, LTD. No. 1:23-cv-00135, at 4 (D. Del. Mar. 29, 2023).↑

  20. Getty Images (US), Inc. v. Stability AI, LTD [2023] EWHC 3090 (Ch) (Eng.), https://www.bailii.org/ew/cases/EWHC/Ch/2023/3090.html (establishing the “Location Issue” as key to the case, placing the burden on the plaintiff to show prima facie evidence that the training and development of Stable Diffusion took place in the United Kingdom).↑

  21. “Against that background, and in summary only, the Defendant contends that its evidence establishes that: (i) all of the computing resources used by the Defendant for its training of Stable Diffusion have, at all times, been located outside the United Kingdom; in particular, all of the training “compute” infrastructure is located in two USA datacentres operated by AWS; (ii) although services have been provided to the Defendant in the United Kingdom by AWS, they have not related to the training or development of Stable Diffusion; and (iii) none of the Defendant’s employees based in the United Kingdom has ever worked on (or had the right equipment to work on) developing or training Stable Diffusion.” Id. para. 56.↑

  22. See Complaint, Getty Images supra note 19.↑

  23. Overview: The TRIPS Agreement, World Trade Organization, https://www.wto.org/english/tratop_e/trips_e/intel2_e.htm (last visited Oct. 29, 2024) (establishing harmonization as one of the main goals of TRIPS).↑

  24. World Trade Organization, Agreement on Trade-Related Aspects of Intellectual Property Rights art. 13, Apr. 15, 1994.↑

  25. Id.↑

  26. See Getty Images supra note 20.↑