Humboldt-Universit?t zu Berlin

This master’s thesis contributes to innovation by analysing the lesser-discussed legal problem arising from?AI development.?The central point of the thesis?is that AI could be held to violate copyright over those works on which it is trained.?This was alleged in the?2022 lawsuit against Copilot, which?became emblematic of the struggle of balancing copyright protection and AI innovation. The Copilot algorithm was trained on Free open-source software (publicly available online), the protection of which is integral for software development.?

?

Legal analysis revealed that in the EU, the text and data mining exception for copyright?will most likely apply, meaning that?Copilot?and many other large language model AIs will have a de facto defence for violating copyright. The US-equivalent exception?is less likely to apply as the factors are evaluated by courts more holistically. Nevertheless, as this is evaluated in court proceedings, there is a significant potential for exploitation of free code by Big Tech. Given that the current legal frameworks permit what is in essence a violation of copyright?representing an existential risk for the software development community, the thesis advocates for a change in the current position.

?

This current regulatory and practical uncertainty also?has the potential to detrimentally affect?an industry already satiated with risk. Thus, lawmakers must either re-evaluate existing legal provisions or AI must be trained on restricted datasets, risking worsening the quality of AI.?The difficulty of balancing the competing policy interests underlies this thesis to demonstrate that the quality and accessibility of free open-source software should be borne in mind by the stakeholders and policymakers.