Opaque AI threatens the foundations of auditing; a graduated reliance regime and redesign of negligence are needed to allow useful, accountable AI in audits while shifting liability toward vendors and auditors according to explainability, materiality, and corroboration.

Main Finding

The article argues that the rise of opaque machine learning systems in audit practice undermines foundational auditing concepts — the assurance model and the negligence-based liability regime — and that neither outright bans nor uncritical acceptance of opaque AI are appropriate. It proposes a graduated algorithmic reliance framework that conditions how deeply auditors may rely on AI outputs on (1) demonstrable explainability of the system, (2) the materiality of the assertion supported, and (3) the auditor’s ability to independently corroborate outputs. It also reconceptualizes auditor negligence around a “competent hybrid auditor” and recommends reallocating liability among audit firms, technology vendors, and audited entities, with special attention to constraints faced by developing economies.

Key Points

• Problem diagnosis
  • AI is moving from peripheral tools to core operational elements of audit work.
  • Opaque ML models challenge the assurance model (auditor’s ability to obtain and evaluate sufficient appropriate evidence) and the traditional negligence-based liability regime.
• Normative stance
  • Neither blanket prohibition of opaque AI nor uncritical reliance on it is defensible.
  • Regulation and standards should be calibrated to the risks posed by opacity, not to AI per se.
• Proposed framework: Graduated algorithmic reliance
  • Permissible reliance depth should be a function of:
    • Explainability/demonstrable interpretability of the AI system.
    • Materiality/significance of the audited assertion supported by the AI.
    • Auditor’s capacity for independent corroboration of AI outputs.
  • The framework yields a sliding scale: more explainability and corroboration permit deeper reliance.
• Liability and standards
  • Reframes negligence to reflect a “competent hybrid auditor” — an auditor skilled at integrating AI tools while exercising professional skepticism and corroboration.
  • Calls for clearer allocation of liability among audit firms, technology vendors, and auditees.
• Regulatory and comparative analysis
  • Draws on IAASB and PCAOB developments, the EU AI Act, and common law auditor negligence to show gaps and possible alignments.
  • Highlights the particular vulnerability of developing economies where regulatory and cognitive capacity to oversee AI in audits is constrained.

Data & Methods

• Methodological approach: legal and regulatory analysis plus normative argumentation.
• Sources reviewed:
  • Auditing scholarship and professional literature on assurance models and audit evidence.
  • Standard-setting developments and guidance from the International Auditing and Assurance Standards Board (IAASB) and the Public Company Accounting Oversight Board (PCAOB).
  • Comparative regulatory instruments, notably the European Union’s Artificial Intelligence Act.
  • Common-law jurisprudence on auditor negligence and liability allocation.
• Type of evidence: doctrinal analysis of texts, comparative regulatory review, conceptual synthesis; no primary quantitative datasets or empirical testing of audit outcomes are reported.
• Limitations: prescriptive recommendations are grounded in legal/regulatory reasoning rather than microdata on audit performance, cost, or adoption behavior.

Implications for AI Economics

• Adoption incentives and technology markets
  • A graduated reliance regime creates clearer demand signals for explainable AI in audit markets: vendors that can demonstrate explainability and support corroboration will command higher adoption and possibly premium pricing.
  • Vendors supplying opaque “black-box” models may face higher liability risk or restricted use cases, reducing market demand for purely opaque solutions.
• Liability, insurance, and costs
  • Reallocating liability among auditors, vendors, and clients will change insurance pricing for professional liability and product liability, affecting audit firm costs and vendor liability exposure.
  • Compliance costs for auditors will rise where explainability or corroboration requirements are demanding, potentially increasing audit fees and barriers to smaller firms.
• Market structure and competition
  • The “competent hybrid auditor” standard may favor larger firms with resources to integrate, evaluate, and corroborate AI — potentially increasing market concentration unless mitigated by standards or support for smaller firms.
  • Clearer liability rules could enable new third-party services (e.g., explainability providers, independent corroboration services, audit-AI auditors), altering the ecosystem and specialization patterns.
• Information asymmetries and credibility
  • Conditioning reliance on explainability reduces information asymmetries between auditors, clients, and users of audited financials, improving trust in AI-assisted audit outputs and potentially lowering system-level costs of capital if financial reporting quality improves.
• International and development considerations
  • Developing economies with limited regulatory capacity face higher relative costs from implementing the framework; this can slow AI adoption or lead to risky reliance on opaque systems.
  • International standard-setters and multilateral agencies may need to provide capacity-building, shared tooling, or differential regulatory expectations to avoid exacerbating global inequalities in audit quality.
• Policy and research implications
  • Economic research should quantify trade-offs: costs of explainability/corroboration vs. benefits in error reduction, liability mitigation, and market trust.
  • Empirical work is needed on how different liability allocations and explainability thresholds affect vendor pricing, audit market structure, and financial reporting outcomes.