Fordham 33 (Report 7): IP and Frontier Technologies

Key Takeaways

• AI firms prioritize copyright compliance over patent strategy
• US light‑touch, EU rights‑based, China data‑localisation model
• EU's DSM opt‑out training exception proves ineffective in practice
• German court finds AI model contains copyrighted copies; UK court disagrees
• Potential EU compulsory licensing could become default for AI‑generated content

Pulse Analysis

The Fordham 33 panel underscored a pivotal shift in how innovators approach intellectual‑property protection. While early AI debates focused on patents, today’s developers—exemplified by OpenAI’s representatives—spend more resources navigating a patchwork of copyright regimes. The United States continues to favor a laissez‑faire stance, reflecting political gridlock, whereas the European Union imposes a rights‑based framework that many label over‑regulatory. China, meanwhile, blends state‑driven data localisation with emerging consumer‑facing rules that demand truthfulness and alignment with government values. This divergent regulatory landscape forces global AI firms to tailor compliance strategies to each jurisdiction, inflating costs and slowing product rollout.

In Europe, the conversation turned to the practical failures of the DSM Directive’s opt‑out training exception and the broad language of the AI Act’s Article 53. Courts are already testing the limits: Munich’s GEMA v. OpenAI decision found copyrighted material embedded in the model, while England’s High Court in Getty v. Stability AI reached the opposite conclusion. These contradictory rulings highlight the difficulty of applying traditional copyright concepts to machine‑learning outputs. Panelists anticipate that the European Commission may ultimately endorse a collective‑licensing or even compulsory‑licensing scheme, treating AI‑generated content as a public‑domain reservoir of creative heritage rather than a series of isolated infringements.

Beyond immediate compliance, the panel warned of two systemic challenges. First, the extraterritorial nature of AI training undermines the territorial assumptions of existing copyright treaties, creating a looming conflict‑of‑laws crisis as models are trained abroad and deployed worldwide. Second, the financial burden of meeting divergent regulations could entrench market power among a few large incumbents, raising antitrust concerns that regulators have yet to address. As legislators grapple with these issues, stakeholders—from publishers to AI startups—must monitor policy developments closely, because the eventual regulatory equilibrium will dictate the pace of innovation and the distribution of value across the digital economy.