Can Artificial Intelligence Enhance Corporate Financial Risk-Taking Capacity? A Perspective on Innovation Resilience and the Environment

Abstract

In the current global competition, innovation-driven strategies are considered crucial to enhance corporate productivity. Financial risk-taking serves as the baseline for corporate survival. This study scrutinizes the consequences of artificial intelligence on corporate financial risk-taking capacity and elucidates the pathways and mechanisms involved. Using data on local publicly traded entities for the period spanning 2015–2024, this study employs text mining methods and fixed-effects regression analysis to investigate the influence of artificial intelligence on corporate financial risk-taking capacity. The outcomes suggest that AI advances corporate financial risk-taking capacity; specifically, it improves corporate innovation and strengthens innovation resilience (i.e., stability dimension). Furthermore, environmental uncertainty suppresses the constructive influence of AI on monetary risk-taking, whereas high-level environmental information disclosure exerts a positive impact. This study uncovers the underlying processes through which machine learning boosts business financial risk-taking capacity and provides theoretical and practical insights for balancing innovation and risk during corporate digital transformation. In summary, this study makes three key contributions: First, it develops a novel theoretical chain linking AI adoption to enhanced corporate financial risk-taking through the mediating mechanism of innovation resilience. Second, it reveals that the positive effect of AI is attenuated by environmental uncertainty but amplified by environmental information disclosure, integrating external factors into the framework. These findings offer strategic insights for managers and policymakers in the digital era.

1. Introduction

Amid growing global tech rivalry, cutting-edge technology skills have emerged as a major push factor for economic growth and development [1,2]. Promoting integrated and clustered development of strategic emerging industries has created a new wave of growth engines, including next-generation information technologies and artificial intelligence (AI) [3]. A pivotal innovation, AI is driving the current phase of technical upheaval and manufacturing shift, seemingly reshaping industrial structures [4,5], business models, and competitive landscapes with unprecedented breadth and depth. For enterprises, contingent on whether they can seize this wave of AI technology, it is a matter of efficiency in the short term and of innovation capability and sustainability in the long run [6,7,8]. However, AI-driven innovation activities involve substantial capital investments, lengthy processes [9], and significant uncertainties, which pose considerable challenges to an enterprise’s ability to manage risks and sustain innovation [10,11].

Innovation processes are characterized by complexity and dynamism. Conventional research on corporate innovation has primarily focused on examining the efficiency of innovation inputs and outputs. However, as technological iterations accelerate and market fluctuations intensify, mere innovation capabilities are not sufficient to ensure competitive advantage. It is necessary to develop the capacity to withstand shocks, adapt to change, and constantly renew innovation—a concept referred to as innovation resilience [12,13]. This concept transcends a company’s potential to merely resist risks and respond to external shocks or internal setbacks for rapid resource reallocation, optimization of innovation strategies, and preservation of core R&D capabilities to not only recover but also enhance innovation activities [14]. It serves as an important basis for enterprises to overcome technology downturns and achieve sustainability [15,16]. Specifically, when AI technology changes constantly, creating robust innovation resilience helps companies to leverage technology opportunities, reduce hazards related to creative shifts, and continue inventive work.

In the current complex and volatile global competitive landscape, the innovation process is characterized by dynamism and uncertainty. Traditional research on corporate innovation has emphasized the input–output efficiency indicators of innovation. However, with the accelerating pace of technological iterations and intense market fluctuations, conventional innovation capabilities can no longer ensure a competitive edge [17,18]. Therefore, “innovation resilience” is a dynamic approach and a core competency to withstand shocks, adapt to changes, and continuously drive innovation. This concept transcends mere risk resistance and emphasizes an enterprise’s ability to rapidly adjust resource allocation, optimize innovation strategies, and sustain core R&D capabilities in the face of drastic external changes or internal setbacks, thereby achieving continuous recovery and iterative advancement of innovation activities [19]. Robust innovation resilience is crucial for enterprises to seize technological opportunities, reduce the risk of innovation disruption, and maintain innovation agility. Simultaneously, entrepreneurial “environmental uncertainty” has intensified [20,21].

Factors such as policy adjustments, fluctuating market demand, and supply chain instability are posing greater challenges to corporate innovation as they assess entrepreneurial ability to manage risks and adjust strategies. Furthermore, with the growing emphasis on sustainable development, ecological data release—functioning in an essential business system and sharing this with external stakeholders—not only influences corporate green reputation and financing capabilities but is emerging as a key element in shaping innovation legitimacy and securing policy support. Therefore, in the innovation paradigm shift, driven by AI, understanding and enhancing innovation resilience, addressing environmental uncertainty, and improving environmental information disclosure are essential for enterprises to pursue high-quality development amid high risk.

Based on the critiques received, this study aims to address several pivotal gaps in the existing literature. Our primary research objectives are twofold: first, to empirically investigate the direct effect of artificial intelligence (AI) adoption on corporate financial risk-taking capacity; and second, to uncover the underlying mediating mechanism through innovation resilience (including its stability and growth dimensions) and the moderating roles of external environmental factors, namely environmental uncertainty and environmental information disclosure. The core research question we address is as follows: How, and under what boundary conditions, does AI enhance a firm’s capacity for financial risk-taking? To answer this, we propose and test an integrated theoretical framework. The key innovations of this paper lie in (1) constructing and validating the complete “AI → innovation resilience → financial risk-taking” transmission chain, moving beyond simple direct-effect analyses; (2) simultaneously examining the dual dimensions of innovation resilience (stability and growth) as mediators, providing a more nuanced mechanism; and (3) introducing environmental uncertainty and disclosure as crucial contextual moderators, thereby enriching the contingency perspective of AI’s impact. These contributions collectively offer a systematic explanation for how firms can balance innovation incentives and risk management during digital transformation. The remainder of the paper is structured as follows. Section 2 reviews the relevant literature and develops hypotheses. Section 3 details the research design and variable construction. Section 4 presents the empirical results and robustness checks. Section 5 provides further discussion, and Section 6 concludes with implications and limitations.

2. Literature Review

Based on a review and synthesis of the existing literature, the theoretical foundation of this paper lies at the intersection of technological innovation theory, corporate risk-taking theory, and environmental strategic management theory [2,5]. Recent discussions on the application of artificial intelligence (AI) in business operations widely acknowledge that AI can foster corporate innovation by optimizing decision-making processes, enhancing resource allocation efficiency, and improving predictive capabilities [22,23]. However, most current studies concentrate on the direct impact of AI on innovation output or operational efficiency, with limited exploration of how AI influences strategic corporate financial risk-taking by shaping internal organizational capabilities. Simultaneously, a paradigm shift is occurring in innovation research, moving from an emphasis on “innovation efficiency” toward a focus on “innovation resilience.” Grounded in this paper’s integrated theoretical framework, innovation resilience is defined as a dynamic organizational capability that enables firms to sustain innovation activity stability and achieve long-term growth amid turbulence, thereby directly linking the firm’s innovation strategy to its strategic risk posture. Regarding risk-taking, whereas traditional studies have centered on financial indicator volatility or R&D intensity, contemporary research—aligning with corporate risk-taking theory—increasingly examines a firm’s risk-taking capacity: the absorptive and buffering capacity inherent in its financial foundation to withstand potential shocks [24]. At the external environmental level, the roles of environmental uncertainty and environmental information disclosure are receiving increasing attention. Studies show that environmental uncertainty exacerbates corporate resource constraints and encourages conservative strategies, whereas high-quality environmental information disclosure can improve corporate financing conditions by reducing information asymmetry [25,26]. Notably, emerging research is exploring the synergistic effects of digital technologies—such as AI, blockchain, and cloud computing—on green finance and ESG performance, providing valuable insights into the interplay between technology, environment, and financial decision-making [14,18]. In summary, while existing research has advanced our understanding of AI applications, innovation resilience, risk-taking, and environmental factors, it has yet to integrate these elements into a unified analytical framework to systematically reveal how AI enhances corporate financial risk-taking capacity by strengthening innovation resilience and how external environmental factors moderate this process. This paper aims to address this theoretical gap.

3. Research Hypotheses

Artificial intelligence reshapes the logic of corporate risk-taking through two parallel pathways. At the resource level, machine learning-based intelligent credit evaluation and dynamic cash flow prediction models can integrate multi-source unstructured data, enabling more accurate pricing for high-risk innovation projects. Digital platforms such as robo-advisor are broadening financing channels, effectively alleviating resource constraints for enterprises engaged in long-term, uncertain innovation activities [7]. At the governance level, AI constructs a dynamic risk management system characterized by “perception–early warning–simulation.” This is achieved through natural language processing for the real-time interpretation of policy and market risks, as well as digital twin technologies for scenario simulations of strategic investments [27,28,29]. This system shifts trial-and-error costs from real-world operations to virtual spaces, significantly reducing the management’s fear of failure and strengthening the board’s institutional courage to adopt exploratory innovation strategies [25].

Therefore, AI provides both the “ammunition” required for risk-taking and the “courage” to accept risks. Based on this, we propose Hypothesis 1:

H1. 

Artificial intelligence enhances corporate financial risk-taking capacity.

The “stability” of technological adaptability denotes the capacity to maintain the continuity and adaptability of innovation activities under external disturbances. AI systems, including Internet of Things detectors and anomaly detection algorithms, enable dynamic monitoring of the entire research and development process, providing timely warnings for potential risks such as supply chain disruptions and deviations in technological pathways [30]. Meanwhile, reinforcement learning-based intelligent decision-making systems can simulate various market scenarios, enabling enterprises to formulate flexible resource allocation plans. These capabilities significantly reduce the likelihood of innovation activities being abruptly interrupted or experiencing severe fluctuations due to external shocks. A stable and predictable innovation process ensures returns on R&D investments in the long term, directly alleviating management’s concerns about wasting innovation investments. Innovation stability encourages enterprises to maintain or even increase R&D investments in cutting-edge areas during economic downturns or market fluctuations, thereby demonstrating stronger financial risk tolerance and risk-taking behavior. Consistent with previous research, this study proposes Hypothesis 2.

H2. 

Artificial intelligence can enhance corporate financial risk-taking capacity by strengthening enterprise innovation resilience and stability.

Innovation resilience is not only reflected in the ability to withstand shocks but, more crucially, in achieving sustained growth and breakthroughs in innovation capabilities within dynamic environments. This is referred to as “innovation resilience (growth dimension).” Artificial intelligence significantly enhances the efficiency and quality of novel output by refining the distribution of research and development resources and accelerating cross-domain knowledge integration. For example, machine learning algorithms can identify high-potential technological directions, while natural language processing and knowledge graph technologies break down information silos and foster breakthrough ideas. Such sustained and efficient innovation growth enables enterprises to continuously accumulate technological barriers and intangible assets, forming a robust “risk buffer.” With innovative achievements and technological leadership, a company gains greater confidence in undertaking long-term, highly uncertain R&D projects. Consequently, by strengthening innovation resilience (growth dimension), AI equips enterprises with the capacity to balance potential financial risks [31,32,33] with successful innovation returns, thereby fostering a greater willingness to take risks in strategic decision-making. Drawing on earlier research, this study proposes Hypothesis 3.

H3. 

Artificial intelligence can enhance corporate financial risk-taking capacity by strengthening the growth dimension of business advancement.

However, the effectiveness of AI is constrained by environmental uncertainties. According to resource scarcity theory, when external environmental volatility intensifies, management tends to concentrate resources on short-term survival, reducing investments in long-cycle innovation. Although AI can enhance risk identification and financing efficiency, rising macro-level risks are likely to weaken its positive impact on corporate financial risk-taking capacity [26,34]. Increased uncertainty drives enterprises to adopt conservative strategies to avoid high-risk R&D initiatives, even when supported by AI. Simultaneously, financial institutions may also tighten credit, further limiting the ability of enterprises to obtain funding through intelligent tools [35,36] and financial risk-taking [37,38]. Therefore, environmental uncertainty is expected to moderate the enhancing effect of AI technologies on corporate financial policy formulation capacity. Derived from the prior, our research posits Hypothesis 4:

H4. 

External instability adversely mediates the impact of automated algorithms on corporate financial risk tendency capacity.

With green transformation gaining global consensus, corporate environmental behavior and transparency are critical factors in resource acquisition [39]. High-quality green knowledge exposure successfully reduces information asymmetry between enterprises and external investors or financial institutions [40,41]. When enterprises provide detailed and reliable disclosure of their environmental responsibility performance, AI-driven financial platforms and investment institutions can access higher-quality data, enabling accurate assessment of the overall operational risks and long-term value of the enterprise [42] rather than focusing solely on environmental risks. Robust environmental information disclosure is high-quality “information fuel” for the financial empowerment capabilities of AI, amplifying the efficiency of resource allocation. This allows external funders to reduce their due diligence costs and risk premiums based on comprehensive information and to improve their willingness to provide financial support to enterprises. Under such circumstances, enterprises not only enhance their intrinsic risk identification and management capabilities through AI technology but also gain superior external financing conditions and legitimacy endorsements owing to their environmental transparency. Thus, this study proposes Hypothesis 5.

H5. 

The degree of ecological transparency favorably mediates the influence of AI on corporate financial risk-taking capacity.

4. Research Design

4.1. Sample Selection

This study uses a sample of Chinese A-share listed firms from 2015 to 2024. Structured data (financial, governance) come from CSMAR, while unstructured data (MD&A text, ESG ratings) are sourced from CNRDS, merged by stock code and year. We extract MD&A text from CNRDS, calculate AI term frequency using a predefined keyword dictionary (e.g., machine learning, deep learning) after text cleaning, and standardize it by total word count. Observations from ST/*ST firms and those with missing key variables are excluded. Continuous variables are winsorized at the 1% and 99% levels, yielding a final sample of 16,911 firm-year observations [40,41].

This study employs quantitative empirical analysis. Using Chinese A-share listed companies from 2015 to 2024 as the sample, we construct a corporate AI application index through text mining techniques and apply fixed-effects panel models for baseline regression. To establish causality, instrumental variable methods address endogeneity concerns. The mediating effect of innovation resilience is tested through stepwise regression, while interaction term models analyze the moderating effects of environmental uncertainty and environmental information disclosure. Robustness checks include alternative variable measurements and adjusted sample periods, with heterogeneity effects examined through subgroup regressions across regions, industries, and technological attributes.

4.2. Variable Definitions

4.2.1. Dependent Variable

Corporate financial risk-taking capacity refers to the financial foundation and buffering space required for a firm to undertake high-risk, high-return projects in its strategic decisions. Drawing on existing research, this study employs the Altman Z-score as the measurement indicator [42]. This metric integrates five core financial ratios—working capital to total assets, retained earnings to total assets, earnings before interest and taxes to total assets, market value of equity to book value of liabilities, and sales to total assets—to comprehensively assess a firm’s financial robustness and risk-bearing capacity. Theoretically, a higher Z-score indicates stronger profit accumulation, a more reasonable capital structure, and more efficient asset operations, thereby providing the necessary financial cushion for firms to engage in high-risk activities such as R&D investment and market expansion [43]. Empirically, the Z-score has been widely validated as an effective predictor of corporate financial distress, with its value inherently reflecting the boundaries of a firm’s financial risk-taking [21,22]. Compared to other indicators, the Z-score considers not only short-term solvency but also long-term profit accumulation and operational efficiency, making it better suited to capture the financial foundation required for sustained risk-taking. Thus, it serves as an ideal proxy variable for measuring corporate financial risk-taking capacity [23].

4.2.2. Independent Variable

In this study, artificial intelligence is the independent variable. This study employs the occurrence rate of machine learning-associated terms within a corporate exchange with evaluation (MD&A) segment as a substitute variable to measure the extent of corporate AI application. This approach is justified both theoretically and practically. From the perspective of informational validity, the MD&A, as a core component of listed companies’ annual reports [44,45], directly reflects management’s authoritative articulation of strategic priorities and technological focus. The frequency of references to AI technologies objectively indicates a company’s actuial investment and emphasis in this field. From a methodological perspective, text mining-based term-frequency analysis mitigates the subjective biases inherent in traditional surveys and the lag associated with patent data. By constructing a specialized AI dictionary containing terms such as “machine learning,” “deep learning,” and “intelligent algorithms,” this method accurately captures a company’s engagement in core technology domains such as natural language processing and computer vision [46]. Existing research demonstrates that the frequency of AI-related terms in MD&A is significantly positively correlated with objective metrics, including innovation outlays and the share of technical personnel, and that such textual features have predictive power for a company’s future performance. Compared to single indicators such as patents or capital expenditures, term frequency analysis not only reflects the breadth of technology applications but also enables the identification of AI’s strategic positioning through contextual analysis, making it an effective method for measuring corporate AI adoption. Artificial intelligence word frequency is listed in the Appendix A.

4.2.3. Mediating Variables

This study adopts two dimensions for measurement: innovation resilience (stability dimension) and innovation output, which are grounded in solid theoretical foundations and empirical feasibility.

Innovation resilience (i.e., stability dimension) is measured by taking the inverse of the coefficient of variation (standard deviation divided by the mean) of the patent application count. This indicator reflects the stability of the innovation output. Lower volatility suggests that a firm can persist in inventive endeavors amid external contextual shifts, reflecting the disturbance-resistant characteristics of innovation resilience.

Advancement in innovation resilience is assessed using the yearly mean increase ratio of a firm’s patent applications. This metric effectively captures the sustained expansion capacity of the innovation output. A higher growth rate indicates that the firm possesses a stable mechanism for transforming R&D investments into technological achievements, which aligns with the core connotation of growth in innovation resilience [47,48,49].

4.2.4. Moderating Variables

The environmental uncertainty indicator is built upon this investigation. A “five-year revenue fluctuation degree” is evaluated by determining a fraction relative to the typical spread around the average of a company’s operating income during the previous five periods. This metric reflects the stability and predictability of corporate revenue. Higher fluctuation typically indicates greater operational risks and external environmental uncertainty for the company [50,51].

The green data release scale is founded on a previous investigation [52]. This “Sino-Securities ESG Rating” adopts the annual ESG score data released via China’s Sino-Securities Index. This rating comprehensively evaluates the output of an enterprise through green, societal, and management aspects and functions as a significant indicator of corporate sustainability and non-financial performance.

4.2.5. Control Variables

Company size and return on equity (ROE) are included to control for potential effects related to firms’ fundamental characteristics and profitability. Listing age, ownership concentration, and property rights are included to control the influence of business oversight and institutional factors. Furthermore, the cash flow ratio, governing body composition, and the ratio of autonomous board members are included to evaluate the company’s liquidity and governance structure. Additionally, it considers the annual and sector constant effects so that the monetary changes and domain differences do not affect the result. The parameters are listed in

Table 1. The parameters.

Variable Type	Variable Name	Variable Symbols	Measurement Method
Dependent variable	Corporate risk-taking	FR	Altman Z-score
Independent variable	Artificial intelligence	AIT	Frequency of AI-related terms in MD&A
Mediating variable	Innovation stability	Volatility	Inverse of the 3-year rolling standard deviation of independently filed patents
	Innovation growth rate	Growth	Ln(independently filed patents + 1)
Moderating variable	Environmental uncertainty	EU	Five-year fluctuation degree of operating revenue
	Environmental Information disclosure	EDI	Sino-Securities ESG Rating
Control variables	Company size	Size	Natural logarithm of total assets at year-end
	Return on equity (ROE)	Roe	Net profit/average net assets
	Listing age	List Age	Ln(current year − year of listing + 1)
	Shareholding ratio of top three shareholders	Top3	Shareholding of top three shareholders/total shares
	Cash flow ratio	Cash flow	Net cash flow from operating activities/total assets
	Nature of property rights	Soe	State-controlled: 1, otherwise: 0
	Number of directors	Board	Natural logarithm of the number of board members
	Proportion of independent directors	Indep	Proportion of independent directors
	Year fixed effects	Year	Controlled in the model
	Industry fixed effects	Ind	Controlled in the model

.

5. Empirical Model

Drawing on existing research, Model (1) was constructed to test Hypothesis 1:

$${FR}_{i,t}={\alpha }_0+{\alpha }_1{\mathrm{AIT}}_{i,t}+{\displaystyle \sum Control{s}_{i,t}+{\displaystyle \sum {Year}_t+{\displaystyle \sum {Ind}_i+{\epsilon }_{i,t}}}}$$

(1)

First, Model (1) examined the impact of corporate AI on corporate financial risk-taking capacity. In this model, subscript t represents the year, and subscript i represents the enterprise; FR_it denotes the corporate financial risk-taking capacity of enterprise i during period t; AIT_it represents the degree of AI of enterprise i during period t; and Controls_it refers to the control variables. Additionally, this study introduces Year as the year fixed effect term to mitigate the influence of individual differences and annual characteristics on corporate financial risk-taking capacity. The constant coefficient regarding the forecasting framework represents α, furthermore, the random disturbance term is ε_(it). If the coefficient in Model (1) appears favorable and meaningful, it indicates that AI can enhance corporate financial risk-taking capacity. All regressions in this study employ robust standard errors adjusted for heteroskedasticity.

Second, to investigate the intermediary impact of enterprise innovation resilience in the connection between AI and corporate financial risk-taking ability, the empirical models are as follows: in Equations (2) and (3), RESit denotes the innovation stability of enterprise i during period t, whereas GROit represents the innovation growth contribution of organization i within interval t. A significantly positive coefficient of AITit in Model (2) suggests that AI improves the innovation stability of companies. Model (3) incorporates the intermediate factor RESit based on Model (1) to determine if AI has an indirect impact on financial risk-taking (FR) through enhancing innovation stability. If the parameter estimate of RESit in Model (3) is markedly positive, whereas the parameter estimate for AIT_it decreases compared to C₁ in Model (1) or becomes insignificant, it suggests that innovation stability plays a partial or complete bridging function within the sequence whereby AI enhances corporate financial risk-taking capacity. Similarly, if innovation growth (GRO_it) is substituted as the variable, the corresponding mediation test models for the growth pathway are constructed accordingly. In these models, Controls_it represents the control variables, and the annual constant impacts are accounted for. The stochastic disruption factor is denoted as ε_i,t.

$${\mathrm{RES}}_{i,t}={\alpha }_0+{\alpha }_1{\mathrm{AIT}}_{i,t}+{\displaystyle \sum {Controls}_{i,t}+{\displaystyle \sum {Year}_t+{\displaystyle \sum {Ind}_i+{\epsilon }_{i,t}}}}$$

(2)

$${FR}_{i,t}={\alpha }_0+{\alpha }_1{\mathrm{AIT}}_{i,t}+{\alpha }_2{\mathrm{RES}}_{i,t}+{\displaystyle \sum {Controls}_{i,t}+{\displaystyle \sum {Year}_t+{\displaystyle \sum {Ind}_i+{\epsilon }_{i,t}}}}$$

(3)

$${\mathrm{GRO}}_{i,t}={\alpha }_0+{\alpha }_1\mathrm{AIT}i,t+{\displaystyle \sum {Controls}_{i,t}+{\displaystyle \sum {Year}_t+{\displaystyle \sum {Ind}_i+{\epsilon }_{i,t}}}}$$

(4)

$${FR}_{i,t}={\alpha }_0+{\alpha }_1{\mathrm{AIT}}_{i,t}+{\alpha }_2{\mathrm{GRO}}_{i,t}+{\displaystyle \sum {Controls}_{i,t}+{\displaystyle \sum {Year}_t+{\displaystyle \sum {Ind}_i+{\epsilon }_{i,t}}}}$$

(5)

To test Hypotheses 4 and 5, namely, the moderating roles of environmental uncertainty (EU) and environmental disclosure intensity (EDI) on the relationship between AI and corporate financial risk-taking capacity, we constructed empirical Models (4) and (5). In Model (4), environmental uncertainty (Eu_i,t) and its interaction with AI (AIT_i,t × EU_i,t) are included in the baseline specification. A markedly adverse parameter of the combined variable suggests that higher environmental uncertainty attenuates the beneficial impact of AI on firms’ financial risk-taking, supporting Hypothesis 4. Model (5) analogously introduces environmental disclosure intensity (EDI_i,t) and its interaction with AI (AIT_i,t × EDI_i,t). A significantly positive interaction coefficient implies that high-quality environmental disclosure amplifies AI’s capacity to enhance corporate financial risk-taking capacity, thereby validating Hypothesis 5. Both frameworks encompass the complete range of control variables (Controls_i,t) along with annual constant influences; ε_i,t represents the idiosyncratic error term.

$${FR}_{i,t}={\alpha }_0+{\alpha }_1{\mathrm{AIT}}_{i,t}+{\alpha }_2{\mathrm{EU}}_{i,t}+{\alpha }_3\mathrm{AIT}\times {\mathrm{EU}}_{i,t}+{\displaystyle \sum {Controls}_{i,t}+{\displaystyle \sum {Year}_t+{\displaystyle \sum {Ind}_i+{\epsilon }_{i,t}}}}$$

(6)

$${FR}_{i,t}={\alpha }_0+{\alpha }_1{\mathrm{AIT}}_{i,t}+{\alpha }_2{\mathrm{EDI}}_{i,t}+{\alpha }_3\mathrm{AIT}\times {\mathrm{EDI}}_{i,t}+{\displaystyle \sum {Controls}_{i,t}+{\displaystyle \sum {Year}_t+{\displaystyle \sum {Ind}_i+{\epsilon }_{i,t}}}}$$

(7)

6. Empirical Results

6.1. Descriptive Statistics

Table 2. Descriptive statistics.

VarName	Obs	Mean	SD	Min	Max	Median
FR	16,911	4.9727	5.7323	−1.4581	66.4352	3.2082
AIT	16,911	1.8838	5.4468	0.0000	53.0000	0.0000
RES	16,911	0.3496	0.4292	0.0059	1.7320	0.9862
GRO	16,911	1.8171	1.6366	0.0000	6.4552	1.7918
EU	16,910	20.4833	19.0020	1.9688	143.9470	14.9393
EDI	16,877	−13.9843	0.6507	−16.3243	−12.6115	−13.9108
Size	16,911	22.3882	1.2502	19.4777	26.4523	22.2201
ROE	16,911	0.0331	0.1926	−2.1749	0.4179	0.0602
ListAge	16,911	2.3421	0.6101	0.6931	3.4340	2.3979
Top3	16,911	0.4576	0.1471	0.1495	0.8640	0.4474
CashFlow	16,911	0.0497	0.0645	−0.1948	0.2656	0.0471
Board	16,911	2.0983	0.1925	1.6094	2.7081	2.1972
SOE	16,911	0.3020	0.4591	0.0000	1.0000	0.0000
Indep	16,911	0.3791	0.0544	0.2857	0.6000	0.3636

presents the summary metrics. The arithmetic mean (standard deviation) of the corporate financial risk Z-score is 4.9727 (5.7323). The AIT variable’s mean is 1.8838, standard deviation is 5.4468, min is 0, and max is 53; the wide range indicates pronounced heterogeneity in Chinese firms’ attention to AI adoption. The distributional features pertaining to the primary and regulatory factors are aligned with those of prior studies. Among the controls, mean firm size is 22.388, implying that sample firms are relatively large; mean ROE is 0.033, denoting a modest average profitability, while its standard deviation of 0.193 reveals substantial cross-firm dispersion; mean listing age is 2.342, indicating that most firms have been listed for a considerable period; mean ownership concentration is 45.76%, pointing to a relatively concentrated equity structure; mean cash flow ratio is 4.97%, in line with typical corporate cash flow levels; mean board size is about eight directors, satisfying the Company Law requirement; and mean state ownership dummy is 0.302, showing that roughly 30% of the sample are state-owned enterprises. For all variables, the midpoint remains proximate to the average, evidencing symmetric distributions that meet the requirements for subsequent empirical analysis.

6.2. Correlation Analysis

We conducted a correlation test to screen the variables and detect potential multicollinearity.

Table 3. Correlation analysis.

	FR	AIT	Size	ROE	ListAge	Top3	CashFlow	Board	SOE	Indep
FR	1
AIT	0.046 ***	1
Size	−0.336 ***	−0.035 ***	1
ROE	0.190 ***	−0.041 ***	0.131 ***	1
ListAge	−0.185 ***	−0.052 ***	0.418 ***	−0.043 ***	1
Top3	0.025 ***	−0.134 ***	0.216 ***	0.167 ***	−0.138 ***	1
CashFlow	0.177 ***	−0.066 ***	0.098 ***	0.300 ***	−0.00900	0.150 ***	1
Board	−0.096 ***	−0.059 ***	0.265 ***	0.052 ***	0.194 ***	0.046 ***	0.032 ***	1
SOE	−0.143 ***	−0.058 ***	0.371 ***	0.014 *	0.457 ***	0.178 ***	−0.021 ***	0.279 ***	1
Indep	0.0120	0.041 ***	−0.022 ***	−0.017 **	−0.046 ***	0.034 ***	0.00400	−0.588 ***	−0.064 ***	1

Note: ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

presents the linkage grid. The primary response factor, corporate financial risk, demonstrates a meaningfully favorable association with the autonomous factor AI, offering early support for Theory 1. Additionally, the total magnitude of every coefficient in the matrix is below 0.6, which is well under the customary threshold of 0.8, indicating the absence of strong correlations and, hence, no serious multicollinearity concerns.

6.3. Main Effects Tests

This study initially scrutinized the association between machine intellect and corporate financial risk-taking capacity based on Model (1).

Table 4. Regression results of main effects.

	(1)	(2)	(3)
	FR	FR	FR
AIT	0.048 ***	0.061 ***	0.038 ***
	(5.935)	(8.243)	(4.708)
Size		−1.753 ***	−1.616 ***
		(−46.048)	(−41.633)
ROE		5.552 ***	4.956 ***
		(25.446)	(23.376)
ListAge		0.004	0.107
		(0.049)	(1.321)
Top3		2.464 ***	3.091 ***
		(8.245)	(10.477)
CashFlow		13.615 ***	12.611 ***
		(21.006)	(19.597)
Board		−0.288	−0.486 *
		(−1.053)	(−1.833)
SOE		−0.086	−0.312 ***
		(−0.829)	(−3.030)
Indep		−0.443	−0.923
		(−0.484)	(−1.044)
_cons	4.882 ***	42.904 ***	40.091 ***
	(104.783)	(42.407)	(39.511)
N	16,911.000	16,911.000	16,911.000
r2_a	0.002	0.197	0.263
Ind	No	No	Yes
year	No	No	Yes

Note: *** and * indicate significance at the 1% and 5% levels, respectively.

presents the analytical findings of Model (1). To examine how AI directly affects companies’ willingness to take financial risks, part (1) shows that the analytical measure the money-risk-taking ability of businesses is 0.048, which is much greater than zero at the 1% level of significance. Column (2) has control variables, and the regression coefficient for corporate financial risk-taking capacity is 0.061, which is markedly positive at the 1% significance threshold. Additionally, framework (3) integrates annual and sectoral constant influences, with the regression coefficient for corporate financial risk-taking capacity remaining significantly positive at 0.038 at the 1% threshold. These findings validate that AI has a favorable impact on enterprise fiscal risk-acceptance ability. The results in Table 4 demonstrate a significant positive impact of AI, reflecting how enterprises’ application of AI technologies enhances their adaptability to complex market environments and decision-making precision amid accelerating digital transformation, thereby strengthening their willingness to undertake strategic risks.

6.4. Mediation Effect Test

Table 5. Mediation effect regression results.

	(1)	(2)	(3)	(4)
	RES	FR	GRO	FR
AIT	0.001 *	0.037 ***	0.026 ***	0.037 ***
	(1.910)	(4.624)	(11.109)	(4.630)
RES		0.477 ***
		(5.791)
GRO	0.001 *	0.037 ***		0.019 *
	(1.910)	(4.624)		(0.717)
Size	−0.353 ***	−1.446 ***	0.287 ***	−1.621 ***
	(−97.129)	(−29.778)	(25.145)	(−41.009)
ROE	0.388 ***	4.773 ***	0.501 ***	4.947 ***
	(19.542)	(22.252)	(8.036)	(23.287)
ListAge	−0.068 ***	0.135 *	−0.449 ***	0.115
	(−9.043)	(1.666)	(−18.902)	(1.411)
Top3	−0.124 ***	3.135 ***	−0.205 **	3.095 ***
	(−4.491)	(10.607)	(−2.357)	(10.488)
CashFlow	0.048	12.583 ***	1.650 ***	12.580 ***
	(0.799)	(19.544)	(8.714)	(19.505)
Board	0.052 **	−0.511 *	0.546 ***	−0.496 *
	(2.074)	(−1.924)	(7.003)	(−1.869)
SOE	0.094 ***	−0.363 ***	0.152 ***	−0.315 ***
	(9.807)	(−3.515)	(5.035)	(−3.056)
Indep	−0.009	−0.920	0.897 ***	−0.940
	(−0.104)	(−1.040)	(3.445)	(−1.062)
_cons	−5.909 ***	42.886 ***	−5.142 ***	40.188 ***
	(−62.203)	(38.103)	(−17.220)	(39.261)
N	16,911.000	16,911.000	16,911.000	16,911.000
r2_a	0.514	0.265	0.238	0.263
Ind	Y	Y	Y	Y
year	Y	Y	Y	Y

Note: ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

shows the influence of enterprise adaptive creativity on the association between AI and corporate financial risk-taking capacity. The coefficients for innovation stability are 0.001 in column (1) and 0.037 within segment (2), each favorable at the 10% significance level. This finding indicates that AI technology enhances corporate financial risk-taking capacity by reducing the volatility of innovation activities. AI-driven predictive analytics and intelligent decision-making systems effectively mitigate external environmental shocks to corporate innovation activities, enabling companies to maintain the stability and continuity of their R&D investments. This innovation enhancement provides a solid foundation for enterprises to undertake higher financial risks. In column (3), the coefficient for innovation growth is 0.026, and in column (4), it is 0.037, both within segment (2), and each are substantially favorable in terms of demonstrating that AI also enhances the willingness to take financial risks by fostering continuous expansion of business creativity output. AI technologies optimize R&D processes, accelerate knowledge integration, and expand innovation boundaries, thereby improving the efficiency and quality of patent output. This continuous accumulation of innovation capabilities increases corporate confidence in making high-risk, high-reward investment decisions.

AI enhances corporate financial risk-taking capacity through two pathways: improving innovation stability (reducing innovation volatility) and fostering innovation growth. This aligns with China’s current policy direction of promoting a virtuous “technology–industry–finance” cycle: by leveraging AI to build resilient innovation systems, enterprises can better buffer external shocks while continuously developing core competencies, thereby more proactively seizing strategic investment opportunities during industrial upgrading.

6.5. Moderation Effects

This study uses the last five-year adjusted volatility of operating revenue as a proxy for environmental uncertainty. From the results of the model, it can be observed that the joint component between AI and environmental uncertainty (AIT × EU) is −0.003, which is significantly negative at the 1% level. This finding suggests that a higher degree of environmental uncertainty impedes the positive effect of AI on companies’ willingness to take financial risks. In column (2) of

Table 6. Regression results of moderating effects.

	(1)	(2)
	FR	FR
AIT	0.054 ***	0.627 ***
	(4.481)	(3.465)
EU	−0.013 ***
	(−5.037)
EU_AIT	−0.003 ***
	(−4.098)
EDI		2.088 ***
		(29.089)
EDI_AIT		0.042 ***
		(3.281)
_cons	14.140 ***	37.552 ***
	(16.823)	(32.242)
N	16,911.000	16,911.000
r2_a	0.190	0.234
Control	Y	Y
Ind	Y	Y
year	Y	Y

Note: *** indicate significance at the 1% level.

, the coefficient of the relationship variable between machine learning and environmental information disclosure (AIT × EDI) equals 0.042, which is favorable at the 1% level. This implies that effective environmental information disclosure enhances the machine learning function in developing business financial risk-taking capacity, thereby supporting enterprises in undertaking high-risk investments that are oriented toward green innovation.

6.6. Robustness Tests

6.6.1. Ordinary Least Squares (OLS)

In this study, the explanatory variable is corporate artificial intelligence. Although AI is often considered exogenous from a theoretical perspective, in the empirical analyses, researchers must be mindful of potential endogeneity issues and take appropriate control or correction measures to safeguard the precision and dependability of the statistical analysis outcomes. This study used the Internet penetration rate as an instrument factor and employed a two-step regression technique. After adding the instrumental variable, the reanalyzed findings, which are presented in column (2) of

Table 7. Robustness tests.

Var	(1)	(2)	(3)	(4)
	WEB	FR	SA	FR
AIT_hat		0.004 ***
		(3.309)
AIT	0.214 ***		0.182 ***
	(4.235)		(3.251)
Fin				0.192 ***
				(9.441)
Control	Y	Y	Y	Y
N	16,911.000	16,911.000	16,911.000	12,954.000
r2_a	0.540	0.060	0.063	0.101
year	Y	Y	Y	Y

Note: *** indicate significance at the 1% level.

, remain statistically relevant [53,54].

6.6.2. Alternative Measurement of the Dependent Variable

The study substitutes the dependent variable with the SA index, which is a stand-in for corporate financing limitations, and conducted the regression anew. Column (3) of Table 7 presents the meaningful findings.

6.6.3. Narrowing of the Sample Period

The COVID-19 pandemic in 2020 may have compelled companies to reduce their R&D investment ratios owing to financial problems. Therefore, all samples after 2020 were excluded to avoid any impact on our study’s findings. Thus, the sample period was 2015–2019, and the regression was repeated. As shown in column (4) of Table 7, the results are similar to those of the main regression.

7. Heterogeneity Analysis

Table 8. Heterogeneity test.

	(1)	(2)	(3)	(4)	(5)	(6)
	Eastern	Central-Western	H-Pollution	L-Pollution	H-Tech	L-Tech
	FR	FR	FR	FR	FR	FR
AIT	0.408 **	0.403	0.061	0.338 *	0.654 ***	0.056
	(1.762)	(1.093)	(0.236)	(1.876)	(2.710)	(0.332)
_cons	−0.472	0.048	1.250 ***	−0.744 ***	−0.529	−0.253
	(−1.600)	(0.108)	(2.980)	(−2.705)	(−1.474)	(−0.867)
N	10026	5697	7312	9236	8694	6987
r2_a	0.717	0.659	0.622	0.721	0.701	0.672
Ind	Y	Y	Y	Y	Y	Y
Control	Y	Y	Y	Y	Y	Y
year	Y	Y	Y	Y	Y	Y

Note: ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

represents the result of heterogeneity analysis. First, as there are sizable differences between the eastern and central-western parts of China regarding monetary bases, banks’ places for storing cash, and how people conduct business together, how well AI helps companies take risks with their money may differ depending on their location. This study conducted grouped regressions based on the registered location of enterprises. In the eastern region, where the digital construction system is well developed and marketization is more advanced, AI can effectively identify and align with the financial governance requirements of enterprises, thereby helping to reduce financing costs [55,56]. Moreover, AI technologies can be readily implemented by relying on mature industrial chains and innovation networks. By contrast, the central-western regions have limited financial resources, weaker technological absorption and transformation capabilities, and limited coverage and penetration of artificial intelligence, making it difficult for AI to significantly enhance corporate financial risk-taking capacity.

Second, this study divides the sample into high- and low-pollution industries for the heterogeneity test. Low-pollution industries usually have a strong R&D foundation and technology accumulation [57]. Through flexible financing solutions and risk mitigation mechanisms, AI can better support the improvement of financial performance in these enterprises. However, high-pollution industries often face severe funding pressures and longer return cycles in their green transformation efforts [58,59]. Considering risk control and profitability, AI may prioritize supporting projects with mature technologies and clear short-term returns, thereby limiting its effectiveness in enhancing the financial governance capabilities of enterprises in such industries.

Third, high-tech enterprises typically possess strong scientific research capabilities and technological absorption and transformation capacities. Artificial intelligence can leverage innovative financial instruments, such as intellectual property pledges and science and technology credit, to provide suitable financial support for high-tech enterprises. By contrast, non-high-tech industries rely largely on traditional technological pathways. While AI can alleviate financing constraints reasonably, its role in enhancing the innovation capabilities of non-high-tech industries may be less pronounced [60,61].

8. Discussion and Conclusions

8.1. Discussion

Although prior studies have acknowledged the beneficial effects of AI on corporate operational efficiency, early risk warning capabilities and innovation efficiency and have explored the role of innovation resilience in turbulent environments [12,13], they have yet to incorporate the complete transmission mechanism of AI—innovation resilience—financial risk-taking into a unified analytical framework. Additionally, a comprehensive analysis of the mediating effects of environmental uncertainty and environmental information disclosure is lacking [23]. This study addresses these gaps by constructing a chained mediation-moderation model. This study systematically examined the internal pathways through which AI enhances corporate financial risk-taking capacity by strengthening the stability and growth dimensions of innovation resilience. Furthermore, it reveals the boundary conditions imposed by external environmental factors, providing an integrated theoretical explanation for understanding risk decision-making [62] and corporate innovation in the digital–intelligent era [63].

This study employs text analysis methods to construct indicators of corporate AI applications and measures innovation resilience from the dual dimensions of stability and growth, thus enhancing the scientific rigor and specificity of the variable measurement. Simultaneously, the contextual adaptability and robustness of the research findings were strengthened by introducing environmental uncertainty and environmental information disclosure as the moderating variables. The research implications suggest that enterprises should integrate AI strategies with innovation resilience development and environmental information disclosure to systematically enhance financial risk-taking. At the policy level, measures such as differentiated support mechanisms and green finance should be implemented to alleviate the technological gap and cultivate a sustainable innovation ecosystem.

8.2. Conclusions and Implications

8.2.1. Research Conclusions

According to the results of the empirical test grounded on the statistics regarding national domestic equity, using data from publicly traded enterprises sourced for the period from 2015 to 2024, it can be seen that AI technology improves enterprise capacity to undertake economic uncertainties. This underlying mechanism is that AI does not directly influence risk decisions but operates by shaping and strengthening the core mediating variable of “innovation resilience.” The specific pathways include, on the one hand, AI’s significant improvement of the growth dimension of corporate innovation output by enhancing innovation procedures and expediting insights synthesis, thereby creating an “earnings buffer” against risks; on the other hand, through intelligent early warning and dynamic simulation, AI effectively mitigates the volatility of innovation activities caused by external shocks, ensuring the stability of the innovation process, thereby alleviating management’s concerns about the uncertainty of innovation investments. Furthermore, this study shows that this positive effect is significantly moderated by external environmental factors: environmental uncertainty suppresses the enabling effect of AI, whereas high-quality environmental information disclosure enhances the positive impact of AI on financial risk-taking by improving external financing conditions and legitimacy for enterprises. This effect is evident in the eastern territories, low-emission sectors, and high-technology enterprises.

8.2.2. Implications

Theoretically, enterprises must move beyond viewing AI as merely a tactical tool to reduce costs and improve efficiency. Instead, they should recognize AI’s core role in building dynamic organizational capabilities and reshaping financial decision-making logic from a strategic perspective. Specifically, management should integrate AI into corporate innovation management and risk governance systems, leveraging the advantages of machine learning algorithms in trend prediction, scenario simulation, and anomaly detection. This will enable intelligent monitoring and dynamic optimization of the entire R&D process. This will not only enhance the stability and continuity of innovation activities, thereby reducing the risk of innovation disruptions caused by external shocks, but it can also identify cross-disciplinary technological integration opportunities through technologies such as knowledge graphs, opening new growth pathways. Consequently, it fundamentally strengthens an enterprise’s “confidence” in coping with complex environments and undertaking strategic financial risks. Simultaneously, enterprises should proactively incorporate environmental, social, and governance (ESG) metrics—particularly premium-grade ecological data transparency—within the synergistic scope of their AI strategies. By transparently and systematically disclosing their investments and achievements in green technology R&D, energy conservation, and emission reduction, enterprises can effectively reduce information asymmetry with external investors and financial institutions, thereby shaping a responsible innovator image. This not only improves green financing conditions and lowers capital costs directly but also earns “legitimacy” recognition from regulatory authorities and society. It establishes a solid foundation for resource assurance and social trust in AI-driven innovation projects that are long-term, uncertain, and highly promising.

Practically, policymaking should reflect precision and systematicity to build a supportive ecosystem for AI-driven corporate innovation. First, policies must address the “digital divide” by implementing different support measures. For groups at a relative disadvantage in terms of technological infrastructure, talent pools, and financial resources, such as enterprises in the core and western areas or pollution-intensive sectors, regional AI industry development funds can be established. Additionally, financial benefits, including extra reductions in innovation costs and accelerated amortization of tangible property, may be provided. Enterprises adopting AI solutions for green transformation should also receive targeted subsidies to promote the equitable distribution and sharing of AI benefits. Second, innovative policy tools should be developed to strengthen incentive synergies. Regulatory bodies can promote mechanisms linking “environmental information disclosure quality” with “access to green financial resources.” For instance, enterprises with excellent ESG ratings can receive priority support for green credit approvals and green bond issuances, thus guiding market funds toward AI-driven green innovation projects. Furthermore, to address the inherent long-term and uncertain nature of innovation activities, governments could explore establishing or subsidizing “AI innovation insurance” products. This would provide risk-sharing for key technological R&D and pilot applications, thereby stabilizing innovation expectations. Eventually, the goal is to form a closely interconnected and mutually reinforcing support ecosystem encompassing “technological iterations, policy guidance, and financial empowerment.” This will systematically enhance Chinese enterprises’ innovation resilience and financial risk-taking in the technological age.

To practically integrate AI into corporate innovation and risk governance, it is recommended that the CEO and the chief digital officer take the lead in establishing a collaborative team spanning “technology, business, and risk control.” The implementation can be divided into three steps: first, build an enterprise-level AI data platform and algorithm library to unify data standards. Second, pilot the integration of AI forecasting and simulation tools into R&D and investment decision-making processes, with initial focus on key areas such as green innovation. Finally, incorporate AI risk monitoring indicators into the internal control system and establish a quarterly evaluation and dynamic optimization mechanism. Concurrently, phased objectives should be set: complete pilot projects in key business scenarios within one year, achieve intelligent coverage of core processes within three years, and form an adaptive, iterative intelligent governance system within five years.

8.2.3. Limitations and Future Directions

This study has some limitations that require further research. First, the measurement of corporate AI adoption depends on text analysis methods. This can reflect the technology focus, but it is difficult to accurately grasp the degree of technology integration and application in a particular business context. In the future, we may use data on corporate AI investments, patents, or case studies to obtain more complex measurements. Second, we examine how innovative resilience works in the middle step. However, corporate risk-taking could also be affected by other factors such as the manager’s mindset and what factors employees believe are important to the company. Future studies could include behavioral corporate finance theory to examine the interaction between these non-technical factors and AI. Finally, the research dataset comprises publicly traded A-share enterprises. Whether this conclusion can be generalized to small- and medium-sized enterprises or non-listed companies needs to be verified. Further studies could increase the sample size and investigate the impact of different types of AI technologies on corporate risk-taking behavior, such as generative AI vs. decision-making AI.