Advancing Responsible AI Innovation A Playbook 2025

Government leaders Key roadblocks organizations encounter from the broader ecosystem Undefined data governance standards for generative AI, challenging data minimization principles Fragmented data governance due to factors such as AI nationalism, the nascent adoption of AI standards and conflicting regulations, affecting data sharing and interoperability Limited incentives and lack of standards in emerging data exchange markets, preventing proper oversight of data quality, provenance tracking and fair compensation Legal ambiguities, generating resistance from companies to share data for fear that recipients will use it to train AI models Data market concentration generating information asymmetries and stifling competition, particularly for small- and medium-sized enterprises (SMEs) and start-ups Actions for government leaders –Clarify data governance to account for generative AI: Assess the impact of generative AI on how businesses are incentivized to collect, retain, use and monetize data. Identify and address gaps in current data governance and content management policies. Consider affordances needed for vulnerable or marginalized populations, such as protecting indigenous data sovereignty rights23 or children’s rights24 (see Play 7). –Promote open and inclusive data ecosystems: Develop and harmonize policy and regulatory frameworks to enable responsible data sharing, including legal definitions and guidelines for emerging models like data trusts and cooperatives. For example, the EU’s Data Governance Act supports trusted data intermediaries and promotes data altruism, laying the groundwork for new stewardship and sharing models.25 Communicate legal clarity regarding when data can be used to train models to incentivize sharing without fear of exploitation. In the US, the AI Action Plan directs the National Science Foundation and Department of Energy to create secure compute environments for controlled AI access to restricted federal data, alongside the creation of an online portal for a demonstration project.26 –Enable secure sharing through: –Experimentation support: Establish regulatory sandboxes to test sharing models without facing compliance risks and provide compliance-by-design tooling. –Mutually beneficial data-sharing markets: Promote conditions of clear rules on use, value measurement, contributor rights and compensation, including for aggregators and individuals. Financial markets offer a proven blueprint: just as analysts evaluate stocks and shareholders earn dividends, data markets could employ analysts to assess data quality while compensating data owners for their contributions. –Shared data infrastructure: Facilitate secure, ethical and sovereignty-respecting access to high-quality domestic and cross-border datasets (see Case study 3). –Address synthetic data macro-challenges: While synthetic data offers an alternative to data scarcity, it requires addressing challenges: incentivizing foundation models to revise usage policies that prohibit synthetic data production, providing transparency into biases and limitations, and preventing negative externalities following mass adoption (e.g. model collapse).27 Advancing Responsible AI Innovation: A Playbook 13