AI Safety Governance and the Within-Firm Pay Gap

Our initial sample consists of Chinese A-share firms listed on the Shanghai and Shenzhen stock exchanges from 2019 to 2024. The year 2019 is chosen as the starting point because China established a national AI governance framework in that year. The National New Generation Artificial Intelligence Governance Expert Committee released the Governance Principles for the New Generation of Artificial Intelligence: Developing Responsible AI, which marked the shift toward responsible AI governance at the policy level. Before 2019, firms lacked a unified framework for AI governance disclosure, and related information was difficult to compare across firms. The introduction of these principles provides a consistent basis for measuring firms' AISG across years. We construct the AISG measure from disclosures in annual reports, corporate social responsibility reports, patent documents, and media coverage. Financial, corporate governance, and compensation data are obtained from the China Stock Market and Accounting Research database. Table 1 reports the sample selection procedure. Starting from 29,938 firm-year observations, we apply the following screening steps. First, we exclude financial firms because their accounting standards and compensation regulations differ from those of other industries. Second, we exclude firms under special treatment status during the study period owing to abnormal financial conditions. Third, we remove observations without AISG data. Fourth, we remove observations with missing values for other key variables. These steps leave 3,022 firm-year observations. Because the AISG data start in 2019 and the main explanatory variable enters with a one-year lag, observations in 2019 are excluded for lack of lagged values. The final sample contains 2,202 firm-year observations from 2020 to 2024. To reduce the influence of outliers, all continuous variables are winsorized at the 1st and 99th percentiles.