AI exposes what ESG misses: Hidden risks in corporate financial reporting

A new study has found that artificial intelligence (AI) is uncovering deeper financial irregularities within firms, but the role of environmental, social, and governance (ESG) practices in improving reporting transparency remains conditional and inconsistent. The research highlights how machine learning-driven anomaly detection is reshaping financial oversight while challenging assumptions that ESG engagement universally strengthens accounting quality.

Published in the International Journal of Financial Studies, the study titled "Unsupervised Machine Learning-Based Financial Anomalies, ESG, and Accounting Conservatism" investigates how financial anomaly risks interact with ESG performance to influence accounting conservatism across firms listed on the Stock Exchange of Thailand. The analysis integrates ML techniques with traditional accounting models to examine how firms recognize financial losses under varying levels of risk and governance engagement.

Machine learning exposes financial anomalies but reveals weaker reporting discipline

The study places unsupervised ML at the center of modern financial anomaly detection. Using a composite score derived from Isolation Forest, Autoencoders, and One-class Support Vector Machine models, the research identifies irregular patterns in firm-level financial data without relying on pre-labeled fraud cases.

These models analyze key financial ratios such as profitability, liquidity, leverage, and operational efficiency to detect deviations from expected behavior. By combining multiple detection approaches, the system captures a broader spectrum of anomalies, including subtle irregularities that traditional accounting metrics may overlook.

The results show that firms with higher financial anomaly scores consistently exhibit weaker accounting conservatism. This means that such firms are slower or less consistent in recognizing losses compared to gains, a pattern that increases the risk of overstated financial performance and delayed disclosure of negative information.

This finding challenges the conventional expectation that firms facing higher risk would adopt more conservative reporting practices to protect against litigation or reputational damage. Instead, the study indicates that anomaly-prone firms may engage in less disciplined reporting behavior, potentially exacerbating information asymmetry between management and investors.

Delayed loss recognition can distort earnings quality, mislead investors, and weaken trust in financial disclosures. As anomaly detection becomes more sophisticated, the gap between observed risk and reported performance may become more visible, increasing pressure on regulators and auditors to respond.

The study also demonstrates the effectiveness of unsupervised learning in financial monitoring. Unlike rule-based systems, these models adapt to complex, high-dimensional data environments, making them particularly suited for detecting emerging patterns of financial irregularities. However, their effectiveness also underscores the limitations of existing governance systems, which may not fully align with the risks identified by advanced analytics.

ESG acts as a conditional governance tool rather than a universal safeguard

The study analyses ESG performance as a moderating factor in the relationship between financial anomalies and accounting conservatism. Contrary to the widely held belief that ESG engagement improves transparency and reporting quality, the findings reveal a more nuanced and conditional role.

In low-risk environments, ESG engagement appears to substitute for conservative accounting practices. Firms with strong ESG performance may rely on reputational capital to signal credibility, reducing the perceived need for stringent loss recognition. This can lead to less conservative reporting, even when financial conditions remain stable.

In high-risk environments, the impact of ESG becomes more complex. The study finds that ESG engagement can partially mitigate the negative effects of financial anomalies on accounting conservatism in certain models, particularly those based on earnings time-series analysis. However, this effect is not consistent across all measurement approaches.

In accrual-based models, ESG shows mixed effects, sometimes reinforcing conservatism and sometimes weakening it depending on the interaction with financial anomalies. In market-based models, ESG has little to no significant impact on how firms recognize losses, suggesting that investors may not fully incorporate ESG signals into their assessment of financial reporting quality.

This variability highlights the multidimensional nature of accounting conservatism. Different models capture different aspects of reporting behavior, including managerial discretion, market perception, and earnings persistence. ESG influences these dimensions unevenly, reinforcing the idea that it functions as a state-dependent governance mechanism rather than a universal solution.

The study further suggests that ESG engagement can act as both a complement and a substitute for traditional governance structures. In some cases, it enhances transparency and accountability. In others, it provides a reputational buffer that allows firms greater flexibility in reporting, potentially reducing the urgency of conservative accounting practices.

Governance, risk, and reporting interact in a complex, state-dependent system

Accounting conservatism is shaped by the interaction of financial risk and governance mechanisms, rather than by either factor alone. Financial anomalies serve as indicators of heightened information risk, triggering the need for stronger reporting discipline. ESG, in turn, modifies how firms respond to this risk, but its influence depends on context.

The research frames this interaction as a contingent governance system, where internal risk signals and external legitimacy mechanisms jointly determine reporting outcomes. In this framework, financial anomaly risk acts as a situational trigger, ESG functions as a moderating governance layer, and accounting conservatism emerges as the observable outcome.

Empirical results across multiple models reinforce this perspective. Accrual-based measures show the strongest sensitivity to financial anomalies, indicating that managerial discretion plays a key role in how losses are recognized. Earnings time-series models provide additional evidence of delayed loss recognition in high-risk firms, while also showing that ESG can partially offset this effect. Market-based models, however, exhibit weaker responses, suggesting that external market mechanisms may not fully capture the nuances of reporting behavior.

Additional tests show that governance-specific measures can strengthen loss recognition under high-risk conditions, indicating that internal control mechanisms remain critical for ensuring reporting discipline.

For investors, the findings underscore the need for a more integrated approach to risk assessment. ESG scores alone may not provide a reliable indicator of reporting quality, particularly in firms with high financial anomaly risk. Instead, combining ESG metrics with machine learning-based anomaly detection could offer a more accurate view of financial health.

The study recommends that regulators design oversight frameworks that address both financial reporting and ESG disclosures together. Treating these domains separately may overlook important interactions that influence corporate behavior.

The study suggests that corporate managers should carefully manage reputational goals while ensuring openness in reporting. While ESG investments can enhance credibility and reduce monitoring costs, overreliance on reputational capital may weaken financial discipline in high-risk environments.