This study examines the use and consequences of corporate disclosures that are at least partially written by AI. We employ state-of-the-art Large Language Models (LLMs) to detect and quantify AI-generated text in earnings call presentations during 2015-2024. We document that the use of such text is widespread and increasingly prevalent among firms and industries. Relative to human-written text in earnings calls, AI-generated text has more positive sentiment, greater linguistic sophistication, and more logical and tonal consistency. Instrumental variables regressions show that language generated by AI improves the information quality of earnings calls. Indeed, more use of AI-powered disclosure leads to larger stock market responses to earnings releases, tighter bid-ask spreads, and smaller analyst forecast errors. Mechanism tests reveal that these capital-market effects of AI-generated language are driven not by greater readability, but instead by higher information density. Finally, we find that while the use of AI-powered disclosure lowers firms’ IR wage costs, it does not help firms to reduce their litigation risk or obfuscate unfavorable news.

Using decentralized autonomous organizations (DAOs) as a novel laboratory, I study whether shareholder democracy mitigates conflicts of interest between large and small shareholders. Leveraging granular vote-level data, I document that majority shareholders frequently sway voting outcomes against minority participants. With an event study, I document that first occurrences of "swaying majority voting", signaling conflicts of interest in governance, trigger significant drops in negative abnormal returns (-12.8% weekly CARs). First swaying events also alter governance participation,increasing voting power concentration in subsequent proposals. A model rationalizes my findings, highlighting how majority voters balance private benefit extraction against market perceptions about their types.

The literature on artificial intelligence and labor markets focuses on task displacement and production costs. This paper studies a distinct and overlooked consequence: AI adoption erodes the organizational pipelines through which firms develop senior talent. When AI displaces workers from lower-level tasks, it severs the learning-by-doing process that generates human capital as a costless byproduct of production. Human capital formation becomes an explicit investment whose opportunity cost rises with AI productivity, and the firm rationally curtails entry-level hiring. Embedding this mechanism in a multi-firm economy reveals a prisoner's dilemma: individually rational adoption decisions collectively reduce the supply of high-skilled workers, increasing wages and reducing firm value even as AI raises productivity in lower-level tasks.