Julia Chirkova
A significant disconnect exists between the rapid adoption of artificial intelligence in business operations and the robustness of the governance and oversight mechanisms designed to manage its associated risks. A recent survey by EY reveals a critical governance gap, where the implementation of autonomous AI is outpacing established oversight protocols, leaving organizations vulnerable to material negative consequences. While 98% of senior AI executives reported having formal AI governance policies in place, a startling 47% admitted that their organizations had bypassed these processes for urgent AI deployments. This rapid, often unchecked, integration of AI has led to tangible harm, with over one-third (36%) of respondents reporting AI incidents or failures that resulted in financial damage, operational disruptions, or data loss.
For Certified Public Accountants (CPAs) and financial professionals tasked with advising businesses, this trend signals a need to move beyond merely verifying the existence of AI policies or conducting standard risk assessments. The presence of a documented control, while indicative of a review process, does not guarantee an understanding of how a potential weakness within that process could simultaneously impact cybersecurity, regulatory compliance, operational efficiency, and overall financial performance. As AI becomes increasingly embedded across diverse business functions, a more holistic and interconnected approach to risk assessment is imperative. This approach must provide granular visibility into how disparate exposures interact, where they can compound, and crucially, the timeframe available for intervention before consequences become significant and irreversible.
The Blurring Lines of Interconnected Risks
Traditional risk assessment frameworks often compartmentalize potential threats into distinct categories. Cybersecurity risks are typically managed by IT departments, compliance issues fall under the purview of legal or regulatory teams, and operational risks are assessed separately from financial exposures. Each of these domains often operates with its own dedicated owner, unique set of controls, and independent reporting structures.
However, the pervasive nature of AI means that risks are no longer content to remain within these predefined silos. Consider an AI tool designed to process sensitive financial data or to support critical business decision-making. If this tool relies on inaccurate or biased data inputs, the initial problem might be categorized as a data quality issue. This flawed data, if fed into a financial forecasting model, can directly influence management decisions, leading to strategic missteps. Furthermore, if the AI application inadvertently exposes confidential or personal information, it immediately triggers cybersecurity and privacy concerns. If the organization is then unable to provide a transparent explanation for how an automated decision was reached, it can precipitate significant compliance and governance challenges.
This interconnectedness means that a single underlying weakness can manifest as multiple distinct risk events. When these events are assessed in isolation, a critical visibility problem emerges. The assessment fails to illuminate how these seemingly separate events interact, and more importantly, how the consequences of one risk can exponentially amplify another. This dynamic is becoming increasingly relevant as finance leaders assume greater responsibility for technology-driven strategic decisions. Deloitte’s "Finance Trends 2027" research, published in September 2026, highlighted this shift, indicating that 54% of surveyed finance leaders are now spearheading enterprise AI and technology capital-allocation decisions, with 48% overseeing AI and technology spending and associated cost controls.
For CPAs advising these organizations, the risk assessment process can no longer conclude with the identification of an individual AI risk and the assignment of a generic score. The critical next step must be to thoroughly investigate the potential ripple effects of that identified risk: what else can it impact, and how?
Clarifying Risk Ownership: Beyond Process Management
A fundamental question that CPAs should be posing is straightforward yet profound: "Who owns this risk?" This question extends beyond simply identifying who operates a particular AI process. An employee or a department might manage the day-to-day operations of an AI application, but this operational management does not automatically equate to accountability for the inherent risks stemming from its data inputs, decision-making algorithms, or its connections to other vital business functions. When risk ownership is ambiguous, a critical risk can fall through the cracks, eluding complete oversight from technology, finance, compliance, and operations departments, with no single entity possessing a comprehensive view.
This principle of clear ownership is equally applicable to the effectiveness of controls. The mere existence of a documented AI governance policy does not definitively establish whether an organization possesses the capability to detect when that control is being circumvented or is failing in practice. The EY survey provides a stark illustration of this issue. Among respondents utilizing agentic AI (AI agents capable of independent action), a concerning 26% reported that their organizations could not detect unauthorized AI agents operating internally. This raises a fundamental risk management question: how would the organization even become aware of such a breach?
A control that lacks meaningful detection capabilities offers only a limited early warning system. For CPAs, this implies that a comprehensive assessment must scrutinize not only which controls are documented but also what tangible evidence exists to demonstrate that these controls are functioning as intended and effectively mitigating risks.
To expose these critical gaps in AI risk management, several probing questions can be posed:
- What is the ultimate business decision or process that this AI application supports? Understanding the critical downstream impacts is paramount.
- What are the potential consequences if the AI’s output is inaccurate, biased, or compromised? This forces a consideration of the full spectrum of potential failures.
- If a control fails or is bypassed, what are the specific detection mechanisms in place? This moves beyond policy to practical implementation.
- What is the evidence that these detection mechanisms are tested and effective? Proving functionality is key.
- If a weakness is identified, what is the process for escalating and addressing it across all relevant departments? This ensures a coordinated response.
- Are the assumptions underpinning the design of the control universally applied, or could similar weaknesses exist in other AI applications or processes? This addresses the systemic nature of potential failures.
The final question is particularly critical. Identifying and rectifying a single instance of a weakness does not automatically translate to a reduction in the overarching risk. If the underlying control design or a flawed assumption is replicated across multiple processes or AI applications, a localized fix leaves a broader, systemic exposure intact and unaddressed.
Beyond Likelihood and Severity: The Nuances of AI Risk
Traditional risk assessment methodologies commonly rely on two primary factors: likelihood and severity. Likelihood quantifies the probability of a specific risk event occurring, often informed by historical data and past frequencies. Severity, on the other hand, measures the potential magnitude of the impact if that event were to materialize. While both these factors remain indispensable components of risk evaluation, the unique characteristics of AI-related risks can expose limitations in what these two metrics alone can convey.
An event with a statistically low likelihood of occurrence may still warrant significant attention and resources if its potential impact is exceptionally severe. This is particularly true when the organization possesses a limited ability to anticipate such an event, and when the underlying vulnerability is present across multiple interconnected processes. This is where the concepts of predictability and compounding become crucial additions to the risk assessment framework.
Predictability focuses on the strength and clarity of warning signals available to an organization before a risk event unfolds. A risk characterized by weak or entirely absent leading indicators presents a fundamentally different management challenge compared to a risk that provides ample advance notice, allowing for proactive intervention and response.
Compounding, conversely, examines how a particular risk event can interact with other existing exposures, cascade into additional business functions, or significantly amplify its overall consequences. In an AI-enabled process, a single, seemingly minor control weakness could have far-reaching implications, affecting data integrity, the accuracy of financial reporting, user privacy, cybersecurity defenses, operational continuity, and regulatory compliance – all simultaneously. A point-in-time risk score, while useful for initial identification, is insufficient. A truly comprehensive assessment must elucidate not just the initial exposure but also the potential trajectory and ultimate scope of that exposure.
CPAs: Charting the Course Beyond Initial Findings
As CPAs increasingly assume more prominent advisory roles within organizations, their value proposition extends beyond traditional financial assurance. A significant part of this evolving value lies in their ability to help clients understand the intricate connections between financial and non-financial risks. This necessitates looking beyond the immediate, isolated finding.
For instance, if an AI system generates unreliable or biased information, the risk assessment must meticulously identify all critical business decisions that rely on that specific output. If an AI vendor experiences a significant security breach, the assessment must rigorously determine which organizational data, operational processes, and financial activities are dependent on that vendor and therefore exposed. If employees are observed bypassing established AI governance protocols, advisers must probe the underlying reasons for this circumvention, identify precisely which controls were avoided, and critically, ascertain where else similar behaviors might be occurring. The overarching objective is to clearly distinguish between merely remediating a specific finding and genuinely reducing the underlying, systemic risk.
The pervasive integration of AI amplifies the importance of this distinction. AI has the power to seamlessly connect processes that organizations may still be assessing in a fragmented, siloed manner. While cybersecurity, compliance, operations, and financial exposures might have different designated owners, a single risk event can traverse and impact all four domains with alarming speed.
CPAs are uniquely positioned to challenge this outdated, fragmented view of risk. By consistently asking critical questions – such as who truly owns a particular risk, how would the organization detect its manifestation, where might identical weaknesses exist elsewhere in the organization, and what other processes could be affected – they can provide a far clearer and more actionable picture than a static, isolated risk score.
As artificial intelligence becomes more deeply entrenched in both financial and operational decision-making, risk assessment methodologies must evolve in tandem. Organizations that cultivate the capacity to identify how risks interact, recognize meaningful and predictive warning signals, and meticulously trace weaknesses across diverse business functions will be far better equipped to proactively address emerging exposures before they compound into significant financial or operational crises. This proactive, interconnected approach is no longer a luxury but a fundamental necessity in the age of intelligent automation.
