The Engine and the Brake: Digital Technology Risk and Carbon Emission Efficiency—Evidence From Large Language Models
利用大语言模型构建数字技术风险暴露指数，发现其显著降低企业碳排放效率，主要通过干扰生产和挤出绿色创新实现，且数字成熟度低、投资短视、供应链脆弱的企业受损更大。
ABSTRACT Digital transformation is widely hailed as an engine for corporate productivity, yet the potential for technology risks to act as a brake on environmental efficiency remains largely unexplored. Drawing on digital technology risk exposure as the focal variable, this paper investigates how it relates to corporate carbon emission efficiency (CEE). We construct a novel, granular DTRE index by applying an ensemble of large language models, including FinBERT, ChatGLM, Qwen, and DeepSeek, to analyze the MD&A sections of Chinese A‐share listed firms. Our empirical results provide robust evidence that DTRE significantly suppresses CEE. In practical terms, each standard deviation increase in risk exposure reduces baseline carbon productivity by 10%, representing a penalty with tangible consequences. Mechanism analysis reveals that DTRE inhibits CEE through two primary channels: eroding production efficiency through operational disruptions and stifling green innovation by crowding out research and development resources. Moreover, the DTRE penalty varies across firms: Those with weaker digital maturity, shorter investment horizons, or more vulnerable supply chains suffer disproportionately larger carbon productivity losses. Meanwhile, we find that external factors, specifically regional digital infrastructure and strict local environmental regulations, act as buffers that mitigate the adverse impact of DTRE. These findings underscore the digital paradox and suggest that firms must effectively orchestrate digital resources to balance technological advancement with low carbon development goals.