The revolutionary potential of artificial intelligence (AI) creates substantial uncertainty about its impact on firm value. On one hand, AI can drive revenue growth through innovative products and precise customer targeting, while also reducing operational costs by automating tasks, optimizing supply chains, and improving production processes. On the other hand, AI comes with heightened risks, including regulatory uncertainties, cybersecurity vulnerabilities, and potential operational failures.

Given these uncertainties, it is essential for investors to have high-quality information about firms’ AI initiatives and the associated risks. However, the current regulatory framework makes disclosure of such information voluntary. The U.S. Securities and Exchange Commission (SEC) has expressed concerns about the credibility of corporate AI disclosures, warning against AI washing – overstating a company’s AI capabilities to make them appear more advanced.[1] This practice has fueled skepticism among investors, contributing to a rise in securities class action lawsuits. In a new paper, we seek to shed light on the determinants and information content of corporate AI disclosures in annual reports.

Measuring AI disclosure is challenging due to its voluntary nature and the lack of a standardized framework. Firms often distribute AI-related content across multiple sections of their 10-K reports, such as business descriptions, risk factors, and management discussion and analysis (MD&A). To address these measurement challenges, we developed a two-step approach that combines keyword searching and large language model (LLM)-based analysis. We began with a simple search for “artificial intelligence” in the 10-K filings of U.S. public firms listed on the SEC EDGAR database. We then expanded the list of keywords by manually reviewing 10-K reports that discussed AI, identifying commonly occurring terms. Traditional keyword-based searches, however, may miss contextual references to AI or misinterpret the nature of AI discussions. To address this, we leveraged ChatGPT to provide contextual analysis. For every identified AI-related keyword, we extracted 400 words before and after the keyword and asked ChatGPT to classify the nature of the AI mention into five categories: product development, pricing optimization, AI product provision, inventory management, and operational efficiency. This approach allowed us to categorize the nature of AI disclosure more accurately and link AI mentions to specific areas of firm activity.

Our analysis of 10-K filings from 2010 to 2023 reveals a significant increase in the frequency and concentration of AI disclosure. In 2010, only 2.36 percent of firms mentioned AI in their 10-K reports. By 2023, this figure had risen to 20.02 percent. Importantly, the increase occurred across all industries, reflecting the growing importance of AI in sectors like manufacturing, healthcare, finance, and retail. Most AI-related discussions are concentrated in the business description, risk factors, and management discussion and analysis sections of the 10-K report. These sections offer insight into how firms incorporate AI into their operations, assess associated risks, and outline future strategic initiatives.

To understand what motivates firms to disclose AI information, we investigated the relationship between firm characteristics and AI disclosure. We found that firms with a higher proportion of AI-skilled employees are significantly more likely to disclose AI information. Specifically, a 1 percent increase in AI-skilled employees increases the likelihood of AI disclosure by 0.6 percent. Other key determinants include firm size, valuation, and firm age. Larger firms are more likely to disclose AI initiatives due to their greater resources to invest in new technologies. Firms with higher market valuations tend to highlight AI to emphasize their innovation capabilities. Younger firms, often seen as disruptors, are more inclined to disclose AI-related initiatives as part of their growth narrative.

We then explored whether AI disclosures contain useful forward-looking information for investors. By linking AI disclosure to subsequent firm performance, we observed that AI disclosure is positively associated with future sales growth, employment growth, capital investment, and R&D intensity. Importantly, AI disclosures also correlate with increased firm risk, as evidenced by a rise in stock price volatility and option-implied volatility. This suggests that while AI may drive growth, it also introduces uncertainties. By classifying the nature of AI activities into categories such as product development, pricing optimization, inventory management, and operational efficiency, we found that both revenue-enhancing and cost-reducing AI activities significantly affect firm performance. Firms that use AI for both revenue generation and cost reduction exhibit stronger future growth compared with firms focused on only one aspect of AI adoption.

Given the inherent risks of AI adoption, companies often disclose AI-related risks in the risk factors section of their 10-K reports. We classified these risks into six categories: regulatory risks, operational risks, competitive risks, cybersecurity risks, ethical risks, and third-party risks. Regulatory risks stem from uncertainty in AI regulations and compliance requirements, while operational risks arise from integration failures and system malfunctions. Competitive risks are linked to threats from rival firms with superior AI technology. Cybersecurity risks are associated with the potential for data breaches and hacking of AI-driven systems. Ethical risks include concerns about fairness, discrimination, and societal impact, while third-party risks involve reliance on external AI vendors and service providers. Firms that disclose more AI risks exhibit higher stock and option-implied volatility, indicating that markets recognize the uncertainty associated with AI adoption.

Our study highlights the growing importance of corporate AI disclosure in annual reports. We provide evidence that AI disclosures offer valuable forward-looking information related to firm growth, operational efficiency, and risk. The adoption of advanced LLMs like ChatGPT allows for a more precise analysis of disclosure content, shedding light on the specific nature of firms’ AI activities. Our findings have implications for investors, regulators, and policymakers, particularly as regulatory efforts around AI evolve. In light of increasing pressure for transparency, firms may face growing scrutiny over the quality and credibility of their AI disclosures.

ENDNOTE

[1] SEC Chair Gary Gensler, in a September 2024 statement, emphasized the need for companies to ensure that their AI-related claims align with their actual capabilities, signaling the regulator’s increasing focus on this issue. See https://www.thecorporatecounsel.net/blog/2024/09/sec-chair-addresses-ai-washing-by-public-companies.html.

This post comes to us from Yang Cao and Miao Liu at Boston College’s Carroll School of Management, Jiaping Qiu at McMaster University’s Michael G. DeGroote School of Business, and Ran Zhao at San Diego State University. It is based on their recent article, “Information in Disclosing Emerging Technologies: Evidence from AI Disclosure,” available here.