Generative AI and U.S. Financial Reporting Integrity: Detecting Narrative Manipulation, Risk Disclosure Gaming, and Fraud Signals in 10-K Filings

Abstract

U.S. capital markets rely on high-integrity corporate disclosure, yet the narrative portions of annual reports—particularly Management’s Discussion and Analysis (MD&A) and Risk Factors—remain vulnerable to strategic language management and, increasingly, generative-AI-assisted drafting. Unlike traditional misstatement detection that focuses on accounting ratios, narrative manipulation can distort investor beliefs through selective emphasis, obfuscation, boilerplate recycling, and inflated forward-looking language while remaining difficult to audit at scale. This paper develops a conceptual–methodological framework to measure disclosure integrity and detect narrative manipulation in U.S. 10-K filings. We construct a transparent EDGAR pipeline that parses MD&A and Risk Factors and extracts auditable features capturing (i) financial-context tone and uncertainty, (ii) readability and complexity, (iii) risk-factor novelty versus boilerplate drift (year-to-year similarity), (iv) forward-looking intensity, and (v) “tone–fundamentals gaps” that flag potential inconsistency between narrative claims and contemporaneous performance signals. We benchmark interpretable statistical models (logit/elastic net) against flexible machine-learning specifications while preserving explainability, and we evaluate performance using governance-relevant outcomes including enforcement-linked misstatement events and restatement proxies, complemented by market-based reactions around filing dates. We further test whether narrative-risk signals strengthen in the period associated with broad diffusion of generative AI writing tools. The proposed framework contributes to accounting, finance, and regulatory technology by converting narrative disclosure risk into measurable, monitorable indicators. Findings are positioned to inform SEC disclosure oversight, audit planning, and issuer governance by enabling scalable detection of narrative manipulation and risk-disclosure gaming that may undermine investor protection and U.S. market integrity.