China: Adopted AI Safety Governance Framework

On 6 September 2024, the National Information Security Standardization Technical Committee (TC260) adopted the Artificial Intelligence (AI) Safety Governance Framework, establishing guidelines on AI development. The Framework outlines measures to address various types of AI safety risks. It stipulates principles for AI safety governance, including committing to sustainable security, prioritising innovative development of AI, establishing governance mechanisms that engage stakeholders, integrate technology and management, and ensuring that all parties involved fully take on their responsibilities for AI safety. Further, the Framework outlines control measures to address different kinds of safety risks. As such, it sets out an extensive classification of AI safety risks, including but not limited to risks of explainability, risks of bias and discrimination, risks of robustness, risks of stealing and tampering, risks of unreliable output, risks of adversarial attack, risks of illegal collection and use of data, risks of unregulated training data annotation, risks of exploitation through defects and backdoors, and risks of supply chain security. The Framework sets out technological measures to combat each of the identified risks, like constant improvement of explainability and predictability, establishment of security rules on data collection and usage, and strict selection of training data. Lastly, the Framework establishes comprehensive governance measures, spanning from implementing a tiered and category-based management for AI application, strengthening AI supply chain security, to promoting international exchange and cooperation on AI safety governance. The Framework concludes with safety guidelines for AI development and application.