Reshaping the game - 2026 report
The copyright quagmire: training data, ownership, and third-party rights
Courts on both sides of the Atlantic demand human creativity that shows in the output - training data has become the expensive end, ownership the strategic one, and infringement screening the operational one.
Section 3 in four theses
- Training on protected material without a licence carries real litigation risk; the first European decisions and a nine-to-ten-figure US settlement mark the price.
- Purely AI-generated assets enjoy no copyright, in the EU or the US. Human creative decisions must be visible in the output; effort and prompt length count for nothing.
- The burden of proof is shifting: once AI generation is plausibly alleged, the rightsholder must document the creative process.
- How much AI a studio deploys on which assets is a business decision with legal consequences, there are several defensible paths, not one right answer.
As AI adoption accelerates, copyright has emerged as one of the most important legal issues facing the games industry. The core conflict pits the foundational need of AI models for vast amounts of training data against the fundamental principles of IP protection. Studios must therefore consider both the risks associated with the AI tools they use and the ownership and infringement issues that can arise from AI-generated content. Since the first edition of this guide, courts have begun to answer questions that were still theoretical in 2025 - this section integrates those decisions.
Training data
On 31 July 2026 we saw the first European judgment on audio-generating AI (LG München I, 42 O 763/25) and the CJEU's answers in Like Company.
Ownership: the human-authorship standard
Human creative decisions must survive the AI process and show in the output - effort, cost and prompt length count for nothing.
Third-party rights: screening AI output
AI multiplies accidental infringement; screen outputs against the CJEU's two-step test, and do not over-rely on pastiche, adaptation defences or indemnities.
New challenges from agentic AI (real-time content generation)
Agentic AI demands architectural answers across the entire pipeline. In production, high autonomy trades away IP ownership; in live gameplay, dynamic generation requires a dual defence - anchored human assets to retain copyright on one side, and strict guardrails to mitigate real-time liability on the other.
Contacts
© 2026 Bird & Bird