AI Safety Governance and Corporate Innovation Speed

In 2019, the National New Generation Artificial Intelligence Governance Expert Committee of China was established and released Governance Principles for a New Generation of Artificial Intelligence. Developing Responsible Artificial Intelligence, which for the first time incorporated AI safety governance as a core item on the national policy agenda and marked the starting point of China's systematic advancement in this area. To examine firms' voluntary governance practices during the initial establishment and diffusion of this framework, we use Chinese A-share listed companies from 2019 to 2024 as our initial sample. Patent texts are obtained from the IPRDB Intellectual Property Database, annual reports and corporate social responsibility reports are collected from Cninfo , and official news texts and financial data are drawn from the CSMAR database.
The data are screened as follows. We exclude firms classified as Special Treatment during the sample window, financial sector firms, and observations with missing variable values. Our analysis relies on a survival model that requires further sample screening. We define spell length as the span of years since a firm's last R&D investment spike. Following Chen and Jin (2023), we exclude observations prior to each firm's first R&D investment spike. Building on this, and following Billett et al. (2011), we further remove firms meeting any of four additional criteria, namely firms with only one uncensored spell and one censored spell in which the censored spell is shorter than the uncensored spell, firms with only one R&D investment spike, firms with no R&D investment spike, and, for firms with multiple spikes, the first spell length and spike year. After this screening, the final sample comprises 1,903 firm-year observations. All continuous variables are winsorized at the 1 percent level.

2019年，中国国家新一代人工智能治理专业委员会成立，并发布《新一代人工智能治理原则——发展负责任的人工智能》，首次将人工智能安全治理纳入国家政策议程的核心事项，标志着中国在该领域系统化推进的开端。
为考察该治理框架初步建立与推广阶段企业的自愿性治理实践，本文选取2019—2024年中国A股上市公司作为初始研究样本。专利文本取自IPRDB知识产权数据库（IPRDB Intellectual Property Database），年报与企业社会责任报告采集自巨潮资讯网（Cninfo），官方新闻文本与财务数据则来源于CSMAR数据库（CSMAR）。本次研究对样本开展如下筛选流程：剔除样本区间内被实施特别处理（Special Treatment）的企业、金融行业企业，以及存在变量缺失值的观测样本。由于本文的分析依托生存模型，需进一步开展样本筛选工作。我们将“持续期”定义为企业自最近一次研发投入峰值起的年度跨度。参照Chen与Jin（2023）的研究范式，我们剔除每家企业首次研发投入峰值之前的所有观测样本。在此基础上，遵循Billett等人（2011）的研究方法，我们进一步移除符合以下四项额外标准之一的企业：仅存在1个未截尾持续期与1个截尾持续期且截尾持续期短于未截尾持续期的企业、仅发生1次研发投入峰值的企业、未发生任何研发投入峰值的企业，以及存在多次研发峰值的企业的首个持续期与峰值年度样本。经过上述全部筛选流程后，最终样本共包含1903个公司-年度观测值。所有连续变量均在1%水平上进行缩尾处理。