protected subject matter

Protected subject matter refers to intellectual creations safeguarded by copyright and related rights under international treaties like the Berne Convention and national laws. It encompasses not only traditional works (literature, art) but also performances, sound recordings, and broadcasts. In AI risk management, this concept is critical as AI training datasets often contain such materials. The EU's Directive 2019/790, particularly its exceptions for Text and Data Mining (TDM) in Articles 3 and 4, directly addresses this issue to balance innovation and creator rights. It is distinct from 'personal data' governed by GDPR, which protects privacy. While they can coexist in the same data, their legal bases and compliance requirements differ, necessitating separate management in corporate data governance frameworks.

Enterprises can integrate the management of protected subject matter into their AI risk framework in three steps. Step 1: Data Inventory and Classification. Create a comprehensive inventory of AI training data, classifying it by source and rights status (e.g., public domain, licensed, fair use), aligning with ISO/IEC 38505-1 on data governance. Step 2: Rights Clearance and Risk Assessment. For protected data, conduct a legal analysis to determine if a license is needed or if an exception, like the EU's TDM rules, applies. Document this assessment for auditability. Step 3: Compliance Monitoring. Implement tools to monitor data pipelines for potential copyright infringements and establish an incident response plan. A global tech firm applied this process, reducing potential infringement events in their generative AI development by 90% and ensuring a successful compliance audit.

Enterprises, particularly in regions like Taiwan, face three key challenges. First, Legal Ambiguity: Unlike the EU, many jurisdictions lack a clear statutory exception for Text and Data Mining (TDM), creating uncertainty around the legality of using copyrighted data for AI training. The solution is a risk-averse strategy, prioritizing licensed/public domain data and thoroughly documenting any 'fair use' claims. Second, Licensing Complexity and Cost: Acquiring licenses for the vast datasets needed for large models is often operationally and financially unfeasible. Mitigation involves exploring data collaboratives, investing in niche datasets, and leveraging synthetic data. Third, Lack of In-house Expertise: There is a common shortage of professionals with hybrid legal-tech skills for AI governance. The solution is to engage external experts to build an initial framework and upskill internal teams for long-term management.

Winners Consulting specializes in protected subject matter for Taiwan enterprises, delivering compliant management systems within 90 days. Free consultation: https://winners.com.tw/contact